[ { "path": ".gitattributes", "content": "*.ts linguist-language=JavaScript\n" }, { "path": ".gitignore", "content": "## macOS\n.DS_Store\n\n## emacs\n*~\n\\#*\n.\\#*\n\n## vim\n*.swp\n*.swo\n\n## VSCode\n.vscode/\n\n## PROJECT::GENERAL\nlogs/\n*.log\nnpm-debug.log*\nnode_modules/\ntypings/\n.npm\n.eslintcache\n\n## PROJECT::SPECIFIC\ndst/\n\n## Ignore Visual Studio temporary files, build results, and\n## files generated by popular Visual Studio add-ons.\nobj/\n*.dll\n*.dll.config\n*.pdb\n*.suo\n\n# Chutzpah Test files\n_Chutzpah*\n" }, { "path": "LICENSE", "content": "The MIT License (MIT)\n\nCopyright (c) 2017 wellflat\n\nPermission is hereby granted, free of charge, to any person obtaining a copy of\nthis software and associated documentation files (the \"Software\"), to deal in\nthe Software without restriction, including without limitation the rights to\nuse, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of\nthe Software, and to permit persons to whom the Software is furnished to do so,\nsubject to the following conditions:\n\nThe above copyright notice and this permission notice shall be included in all\ncopies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\nIMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS\nFOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR\nCOPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER\nIN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN\nCONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.\n" }, { "path": "README.md", "content": "## Image processing and Machine learning labs   [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg?style=flat-square)](https://opensource.org/licenses/MIT)\n\ncomputer vision, image processing and machine learning on the web browser or node\n\n### note\n\n* Fast Fourier Transform (1D/2D-FFT)\n* Stereo Matching\n* Poisson Image Editing\n* Line Segment Detector\n* Corner Detection\n* Fish-Eye Transform\n* Image Processing Filters\n* Image Histogram Calculation\n* Image Feature Extraction\n* Decision Tree Learning\n* K-Means++ Clustering\n* Logistic Regression\n* Adaptive Regularization of Weight Vectors (AROW)\n* Soft Confidence Weighted Learning (SCW)\n* Gradient Boosting Decision Tree (GBDT)\n* Neural Network (Denoising Autoencoders)\n* t-Distributed Stochastic Neighbor Embedding (t-SNE)\n* 3D Shape Drawing (Mobius Strip, Klein Bottle, Heart Surface ...)\n* WebGL Samples\n* ONNX Runtime for Web (ORT Web)\n* etc..\n\n### demo\n[Demo Site](https://rest-term.com/labs/html5/index.html) \n\n### license\n\nCopyright © 2017 wellflat Licensed under the [MIT License][MIT]\n\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n" }, { "path": "cv/corner_detection/README.md", "content": "# Corner Detection Module\n\n## description\n\nCorner detection using Harris operator\n\nsee also [blog entry][entry]\n\n### sample\n[![corner_detection](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/room_corners.jpg)](http://rest-term.com/labs/html5/corner.html)\n\n(Tests: IE9, Firefox15.0, Chrome21.0, Safari6.0, Opera12.0)\n\n## usage\n\n```js\n// parameters for Harris corner detection\n// blockSize: Neighborhood size (3×3 or 5×5)\n// k: Harris detector free parameter (recommends 0.04 ~ 0.15)\n// qualityLevel: Parameter characterizing the minimal accepted quality of image corners\nvar params = { blockSize: 3, k: 0.04, qualityLevel: 0.01 };\n// img: ImageData object\n// returns Array of detected corners\nvar corners = CornerDetector.detect(img, CornerDetector.HARRIS, params);\n\n```\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[entry]: http://rest-term.com/archives/2986/\n" }, { "path": "cv/corner_detection/cornerdetect.js", "content": "/**\n * Corner Detector module\n */\n(function() {\n var CornerDetector; // top-level namespace\n var _root = this; // reference to 'window' or 'global'\n\n if(typeof exports !== 'undefined') {\n CornerDetector = exports; // for CommonJS\n } else {\n CornerDetector = _root.CornerDetector = {};\n }\n\n var version = {\n release: '0.1.0',\n date: '2012-09'\n };\n CornerDetector.toString = function() {\n return \"version \" + version.release + \", released \" + version.date;\n };\n\n // core operations\n var core = {\n detect : function(img, type, params) {\n core._validateParams(params);\n switch(type) {\n case 'eigen':\n // implement ?\n throw new Error('not implementation error');\n break;\n case 'harris':\n return core._detectHarris(img, params);\n break;\n case 'fast':\n // implement ?\n throw new Error('not implementation error');\n break;\n default:\n return core._detectHarris(img, params);\n break;\n }\n },\n _validateParams : function(params) {\n var _params = ['qualityLevel', 'blockSize', 'k'], msg = '';\n for(var i=0; i<_params.length; i++) {\n if(!params.hasOwnProperty(_params[i])) {\n msg = 'invalid parameters, required \\'' + _params[i] + '\\'';\n throw new Error(msg);\n }\n if(!parseFloat(params[_params[i]])) {\n msg = 'invalid parameters, required number \\'' + _params[i] + '\\'';\n throw new TypeError(msg);\n }\n }\n if(params.blockSize%2 !== 1) {\n throw new Error('odd number required \\'blockSize\\'');\n }\n if(params.blockSize > 5) {\n throw new Error('unsupported \\'blockSize\\' ' + params.blockSize);\n }\n },\n /* Harris Operator */\n _detectHarris : function(img, params) {\n var w = img.width,\n h = img.height,\n imgdata = img.data,\n len = w*h << 2,\n src, cov, eig, corners,\n r = (params.blockSize - 1)/2,\n dx, dy, dxdata, dydata, kernelx, kernely, maxval, quality,\n x, y, kx, ky, i, j, step, tmp;\n\n if(typeof Float32Array === 'function') {\n src = cov = new Float32Array(w*h*3);\n corners = new Float32Array(w*h);\n } else {\n src = cov = corners = [];\n }\n // change container, uint8 to float32\n for(i=0,j=0; i tmp) {\n tmp = 0;\n break;\n }\n }\n }\n if(tmp !== 0) {\n corners[i] = eig[i];\n }\n }\n }\n // threshold\n maxval = 0;\n len = eig.length;\n for(i=0; i maxval) maxval = corners[i];\n }\n quality = maxval*params.qualityLevel;\n for(j=0; j> 2;\n // bit reversal\n for(var l=0; l<_n; l++) {\n m = _bitrev[l];\n if(l < m) {\n tmp = re[l];\n re[l] = re[m];\n re[m] = tmp;\n tmp = im[l];\n im[l] = im[m];\n im[m] = tmp;\n }\n }\n // butterfly operation\n for(var k=1; k<_n; k<<=1) {\n h = 0;\n d = _n/(k << 1);\n for(var j=0; j> 1;\n while(k <= j) {\n j -= k;\n k >>= 1;\n }\n j += k;\n _bitrev[i] = j;\n }\n },\n // makes trigonometiric function table\n _makeCosSinTable : function() {\n var n2 = _n >> 1,\n n4 = _n >> 2,\n n8 = _n >> 3,\n n2p4 = n2 + n4,\n t = Math.sin(Math.PI/_n),\n dc = 2*t*t,\n ds = Math.sqrt(dc*(2 - dc)),\n c = _cstb[n4] = 1,\n s = _cstb[0] = 0;\n t = 2*dc;\n for(var i=1; i> 1;\n for(var y=0; y> 1,\n sqrt = Math.sqrt;\n for(var y=-n2; y> 1,\n sqrt = Math.sqrt;\n for(var y=-n2; y radius) {\n re[p] = im[p] = 0;\n }\n }\n }\n },\n // applies Band-Pass Filter\n BPF : function(re, im, radius, bandwidth) {\n var i = 0,\n p = 0,\n r = 0.0,\n n2 = _n >> 1,\n sqrt = Math.sqrt;\n for(var y=-n2; y (radius + bandwidth)) {\n re[p] = im[p] = 0;\n }\n }\n }\n },\n // windowing using hamming window\n windowing : function(data, inv) {\n var len = data.length,\n pi = Math.PI,\n cos = Math.cos;\n for(var i=0; i max) {\n max = spectrum[i];\n }\n }\n imax = 1/max;\n for(var j=0; j> 1;\n }\n }\n _context.putImageData(_img, 0, 0);\n }\n };\n // aliases (public APIs)\n SpectrumViewer.init = core.init;\n SpectrumViewer.render = core.render;\n}).call(this);\n" }, { "path": "cv/fft/filter.js", "content": "/**\n * Spatial Frequency Filtering\n * High-pass/Low-pass/Band-pass Filter\n * Windowing using hamming window\n */\n(function() {\n var FrequencyFilter; // top-level namespace\n var _root = this; // reference to 'window' or 'global'\n\n if(typeof exports !== 'undefined') {\n FrequencyFilter = exports; // for CommonJS\n } else {\n FrequencyFilter = _root.FrequencyFilter = {};\n }\n\n // core operations\n var _n = 0;\n var core = {\n init : function(n) {\n if(n !== 0 && (n & (n - 1)) === 0) {\n _n = n;\n } else {\n throw new Error(\"init: radix-2 required\");\n }\n },\n // swap quadrant\n swap : function(re, im) {\n var xn, yn, i, j, k, l, tmp,\n len = _n >> 1;\n for(var y=0; y> 1;\n for(var y=-n2; y> 1;\n for(var y=-n2; y radius) {\n re[p] = im[p] = 0;\n }\n }\n }\n },\n // apply Band-Pass Filter\n BPF : function(re, im, radius, bandwidth) {\n var i = 0,\n p = 0,\n r = 0.0,\n n2 = _n >> 1;\n for(var y=-n2; y (radius + bandwidth)) {\n re[p] = im[p] = 0;\n }\n }\n }\n },\n // windowing using hamming window\n windowing : function(data, inv) {\n var len = data.length,\n pi = Math.PI;\n for(var i=0; i 0 ? diff : -diff;\n ok(diff < error, \"fft1d -> ifft1d, Re error: \" + diff);\n diff = im[j] - im2[j];\n diff = diff > 0 ? diff : -diff;\n ok(diff < error, \"fft1d -> ifft1d, Im error: \" + diff);\n }\n // 2D-FFT/IFFT\n re = re2.concat();\n im = im2.concat();\n FFT.fft2d(re, im);\n FFT.ifft2d(re, im);\n for(var j=0; j 0 ? diff : -diff;\n ok(diff < error, \"fft2d -> ifft2d, Re error: \" + diff);\n diff = im[j] - im2[j];\n diff = diff > 0 ? diff : -diff;\n ok(diff < error, \"fft2d -> ifft2d, Im error: \" + diff);\n }\n }, false);\n } catch(e) {\n return;\n }\n});" }, { "path": "cv/fft/spectrum.js", "content": "/**\n * FFT Power Spectrum Viewer\n */\n(function() {\n var SpectrumViewer; // top-level namespace\n var _root = this; // reference to 'window' or 'global'\n\n if(typeof exports !== 'undefined') {\n SpectrumViewer = exports; // for CommonJS\n } else {\n SpectrumViewer = _root.SpectrumViewer = {};\n }\n\n // core operations\n var _context = null,\n _n = 0,\n _img = null,\n _data = null;\n var core = {\n init : function(context) {\n _context = context;\n _n = context.canvas.width,\n _img = context.getImageData(0, 0, _n, _n);\n _data = _img.data;\n },\n // render FFT power spectrum on the Canvas\n render : function(re, im, islog) {\n var val = 0,\n i = 0,\n p = 0,\n spectrum = [],\n max = 1.0,\n imax = 0.0,\n n2 = _n*_n;\n for(var i=0; i max) {\n max = spectrum[i];\n }\n }\n imax = 1/max;\n for(var j=0; j> 1;\n }\n }\n _context.putImageData(_img, 0, 0);\n }\n };\n // aliases (public APIs)\n SpectrumViewer.init = core.init;\n SpectrumViewer.render = core.render;\n}).call(this);\n" }, { "path": "cv/fisheye_transform/README.md", "content": "# Fish-Eye Transform Module\n\n## description\n\nCircular Fish-Eye transform\n\nsee also [blog entry][entry]\n\n### sample\n[![fisheye_transform](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/fisheye_transform.jpg)](http://rest-term.com/labs/html5/fisheye.html)\n\n(Tests: IE9, Firefox16.0, Chrome21.0, Safari6.0, Opera12.0)\n\n## usage\n\n```js\n// img: ImageData object\n// focalLength: focal-length Number (int)\n// radius: circle radius Number (int)\nvar focalLength = 55;\nvar radius = 60;\n\n// returns ImageData object\nvar result = Fisheye.transform(img, focalLength, radius);\n\n```\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[FishEye]: http://rest-term.com/labs/html5/fisheye.html\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[entry]: http://rest-term.com/archives/2991/\n" }, { "path": "cv/fisheye_transform/fisheye.js", "content": "/**\n * Fish-Eye Transform module\n */\n(function() {\n var Fisheye; // top-level namespace\n var _root = this; // reference to 'window' or 'global'\n\n if(typeof exports !== 'undefined') {\n Fisheye = exports; // for CommonJS\n } else {\n Fisheye = _root.Fisheye = {};\n }\n\n var version = {\n release: '0.1.0',\n date: '2012-09'\n };\n Fisheye.toString = function() {\n return \"version \" + version.release + \", released \" + version.date;\n };\n\n // core operations\n var _ctx = document.createElement('canvas').getContext('2d');\n var core = {\n transform : function(img, focalLength, radius) {\n var w = img.width,\n h = img.height,\n src = img.data,\n f = focalLength,\n r = radius,\n buff = [0, 0, 0],\n dstimg, dst,\n sqrt = Math.sqrt,\n round = Math.round,\n cx, cy, dx, dy, nx, ny, d, fd, i, j, step;\n\n _ctx.canvas.width = w;\n _ctx.canvas.height = h;\n _ctx.fillStyle = 'rgba(0, 0, 0, 0)';\n _ctx.fillRect(0, 0, w, h);\n dstimg = _ctx.getImageData(0, 0, w, h);\n dst = dstimg.data;\n cx = w/2;\n cy = h/2;\n for(var y=0; y threshold) {\n div = 1/acc;\n data[i] = buff[0]*div;\n data[i + 1] = buff[1]*div;\n data[i + 2] = buff[2]*div;\n data[i + 3] = 1; // as flag\n }\n }\n //data[i + 3] = 255;\n }\n }\n }\n };\n // public API\n Fisheye.transform = core.transform;\n}).call(this);" }, { "path": "cv/image_filter/README.md", "content": "# Image Processing Filters\n\n## description\n\nImage processing filters\n\n* Convolution Filter\n\n[![image_filter](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/convolution_filter.jpg)](http://rest-term.com/labs/html5/convolution.html)\n\n* Symmetric Nearest Neighbor Filter\n\n[![image_filter](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/snn_filter.jpg)](http://rest-term.com/labs/html5/snn.html)\n\n* Jitter Jilter\n\n[![image_filter](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/jitter_filter.jpg)](http://rest-term.com/labs/html5/jitter.html)\n\n### samples\n[Convolution Filter][Convolution]\n\n[Symmetric Nearest Neighbor Filter][SNN]\n\n[Jitter Filter][Jitter]\n\n(Tests: Firefox3.6, Safari5.0, Chrome10.0, Opera11.60)\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[Convolution]: http://rest-term.com/labs/html5/convolution.html\n[SNN]: http://rest-term.com/labs/html5/snn.html\n[Jitter]: http://rest-term.com/labs/html5/jitter.html\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n" }, { "path": "cv/image_filter/cavin.js", "content": "/**\n * Cavin.js\n * Computer Vision and Image Processing Library for HTML5 Canvas\n */\n(function() {\n /**\n * intialize\n */\n var Cavin; // top-level namespace\n var root = this; // reference to 'window' or 'global'\n var ctx = null; // local reference to CanvasRenderingContext2D\n var imgdata = null; // local reference to ImageData\n\n // local reference to slice/splice\n var slice = Array.prototype.slice,\n splice = Array.prototype.splice;\n\n if(typeof exports !== 'undefined') {\n Cavin = exports; // for CommonJS\n } else {\n Cavin = root.Cavin = {};\n }\n\n // current version of the library\n Cavin.version = {\n release: '0.1.0',\n date: '2012/03',\n };\n Cavin.toString = function() {\n return \"version \" + Cavin.version.release + \", released \" + Cavin.version.date;\n };\n\n // set the js library that will be used for DOM manipulation,\n // by default will use jQuery\n var $ = root.jQuery;\n Cavin.setDomLibrary = function(lib) {\n $ = lib;\n };\n\n /**\n * image I/O modules\n */\n var IO = Cavin.IO = {\n read: function() {\n var args = slice.call(arguments),\n img = new Image(),\n opt = IO._parseOpt(args);\n img.src = opt['path'];\n ctx = document.createElement('canvas').getContext('2d');\n img.addEventListener('load', function(e) {\n ctx.canvas.width = img.width;\n ctx.canvas.height = img.height;\n ctx.drawImage(img, 0, 0);\n imgdata = ctx.getImageData(0, 0, img.width, img.height);\n opt['success'](imgdata);\n }, false);\n img.addEventListener('error', function(e) {\n opt['error']('can\\'t load image: ' + opt['path']);\n }, false)\n },\n write: function(data, target) {\n var ctx;\n if(typeof target === 'string') {\n ctx = document.querySelector(target).getContext('2d');\n } else if(typeof target === 'object') {\n ctx = target; // CanvasRenderingContext2D\n }\n ctx.putImageData(data, 0, 0);\n },\n getImageData: function(context) {\n var w = context.canvas.width,\n h = context.canvas.height;\n return context.getImageData(0, 0, w, h);\n },\n _parseOpt: function(args) {\n var opt = {};\n // {path:string, success:function, error:function}\n if(args.length === 1 && typeof args[0] === 'object') {\n if(typeof args[0]['path'] !== 'string') {\n throw new Error('string object required: path');\n } else {\n opt['path'] = args[0]['path'];\n }\n if(typeof args[0]['success'] != 'function') {\n throw new Error('function object required: success');\n } else {\n opt['success'] = args[0]['success'];\n }\n if(typeof args[0]['error'] != 'function') {\n throw new Error('function object required: error');\n } else {\n opt['error'] = args[0]['error'];\n }\n }\n // path, callback\n if(args.length === 2 && typeof args[0] === 'string') {\n opt['path'] = args[0];\n if(typeof args[1] != 'function') {\n throw new Error('function object required');\n } else {\n opt['success'] = args[1];\n }\n }\n return opt;\n },\n };\n // aliases\n Cavin.get = Cavin.IO.read;\n Cavin.put = Cavin.IO.write;\n\n /**\n * image processing filter modules\n */\n var Filter = Cavin.Filter = {\n /* convolution filter (x*x squire kernel) */\n convolution: function(kernel, divisor, bias) {\n var srcimg = imgdata,\n w = srcimg.width,\n h = srcimg.height,\n srcdata = srcimg.data,\n dstimg = ctx.createImageData(w, h),\n dstdata = dstimg.data,\n div = 1/divisor,\n r = Math.sqrt(kernel.length),\n buff = [0, 0, 0],\n i, j, k, v, px, py, step, kstep;\n if(divisor === 0) {\n throw new Error('division zero');\n }\n if(r%2 != 1) {\n throw new Error('square kernel required');\n }\n r = (r - 1)/2;\n for(var y=0; y 255 ? 255 : v;\n v = buff[1]*div + bias;\n dstdata[i + 1] = v < 0 ? 0 : v > 255 ? 255 : v;\n v = buff[2]*div + bias;\n dstdata[i + 2] = v < 0 ? 0 : v > 255 ? 255 : v;\n dstdata[i + 3] = 255;\n }\n }\n return dstimg;\n },\n /* box blur filter */\n blur: function(radius) {\n var kernel = [],\n size = (2*radius + 1)*(2*radius + 1);\n for(var i=0; i255 ? 255 : value;\n }\n return dstImg;\n };\n // jitter filter\n self.jitter = function(amount) {\n var p = self.constructor.prototype,\n w = p.bitmapData.width,\n h = p.bitmapData.height,\n srcData = p.bitmapData.data,\n dstImg = p.ctx.createImageData(w, h),\n dstData = dstImg.data,\n len = dstData.length,\n dx, dy, step, i, k;\n for(var y=0; y> l & 1) && (src[i + j] === cmp)) {\n c = 0;\n break;\n }\n }\n }\n features[k] += c;\n }\n }\n }\n return features;\n };\n self.stardetect = function(param) {\n // TODO\n };\n};\n\n/* utilities module */\n/* image histogram */\nCV.modules.histogram = function(self) {\n // draw histogram on the canvas\n self.drawHistogram = function(bins, context, style) {\n var h = context.canvas.height,\n max = self.getMax(bins),\n len = bins.length,\n step = context.canvas.width/len,\n k = 2;\n style = style || {};\n context.globalAlpha = style['globalAlpha'] || 0.8;\n context.strokeStyle = style['strokeStyle'] || '#00cc66';\n context.lineWidth = style['lineWidth'] || 4;\n for(var i=0; i sum2) ? minsum/sum1 : minsum/sum2;\n };\n self.addHistogram = function(hist1, hist2) {\n if(hist1.length != hist2.length) {\n throw new Error('invalid histogram pair');\n }\n var newhist = [],\n len = hist1.length;\n for(var i=0; i max) max = bins[i];\n }\n return max;\n };\n};\n\n/* web storage */\nCV.modules.storage = function(self) {\n // thin wrapper\n self.getData = function(key) {\n return JSON.parse(window.localStorage[key]);\n };\n self.setData = function(key, value) {\n window.localStorage[key] = JSON.stringify(value);\n };\n};\n" }, { "path": "cv/lsd/.babelrc", "content": "{\r\n \"presets\": [\"react\", \"es2015\", \"stage-0\"],\r\n \"plugins\": [\r\n ],\r\n \"env\": {\r\n \"development\": {\r\n \"presets\": [\"react-hmre\"],\r\n \"plugins\": [[\"react-transform\", {\r\n \"transforms\": [{\r\n \"transform\": \"react-transform-hmr\",\r\n \"imports\": [\"react\"],\r\n \"locals\": [\"module\"]\r\n }]\r\n }]]\r\n },\r\n \"production\": {\r\n \r\n }\r\n }\r\n}" }, { "path": "cv/lsd/.eslintignore", "content": "webpack.config.js\r\ndst/\r\n" }, { "path": "cv/lsd/.eslintrc.json", "content": "{\n \"env\": {\n \"browser\": true,\n \"node\": true,\n \"commonjs\": true,\n \"es6\": true\n },\n \"extends\": \"eslint:recommended\",\n \"ecmaFeatures\": {\n \"jsx\": true,\n \"modules\": true\n },\n \"parserOptions\": {\n \"sourceType\": \"module\"\n },\n \"rules\": {\n \"indent\": [\n \"error\",\n 4\n ],\n \"linebreak-style\": [\n \"error\",\n \"unix\"\n ],\n \"quotes\": [\n \"warn\",\n \"single\"\n ],\n \"semi\": [\n \"error\",\n \"always\"\n ],\n \"no-unreachable\": [\n \"warn\"\n ],\n \"no-console\":\"off\"\n },\n \"plugins\": [\n \"react\"\n ]\n}" }, { "path": "cv/lsd/README.md", "content": "# Line Segment Detector Module\n\n## description\n\nLine detection using Line Segment Detector ([LSD][IPOL]) algorithm\n\n![lsd](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/lsd_apply.jpg)\n\n\nsee also [blog entry][entry]\n\n### demo\n[Line Segment Detector Demo](http://rest-term.com/labs/html5/lsd/index.html) [(mirror)](https://secret-nether-01.herokuapp.com/lsd/)\n\n[![lsd_demo](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/lsd_demo.jpg)](http://rest-term.com/labs/html5/lsd/index.html)\n\n\n## usage in JavaScript (ES2015)\n```js\nimport LSD from './lsd';\n// input 8bit gray scale image\nimport SampleImage from './sample.jpg';\n\nconst image = new Image();\nimage.src = SampleImage;\nimage.onload = () => {\n const width = image.width;\n const height = image.height;\n const canvas = document.createElement('canvas');\n const context = canvas.getContext('2d');\n canvas.width = width;\n canvas.height = height;\n context.drawImage(image, 0, 0, width, height);\n document.getElementById('content').appendChild(canvas);\n const imageData = context.getImageData(0, 0, width, height);\n const detector = new LSD();\n const lines = detector.detect(imageData);\n console.log('lines: ' + lines.length.toString());\n detector.drawSegments(context, lines);\n};\n```\n\n### React Component\n```js\nimport React from 'react';\nimport ReactDOM from 'react-dom';\nimport MuiThemeProvider from 'material-ui/styles/MuiThemeProvider';\nimport LSDComponent from './lsd_component';\nimport SampleImage from './sample.jpg';\n\nReactDOM.render(\n \n \n ,\n document.getElementById('content')\n);\n```\n\n\nlicense\n----------\nCopyright © 2017 wellflat Licensed under the [MIT License][MIT]\n\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[IPOL]: http://www.ipol.im/pub/art/2012/gjmr-lsd/\n[entry]: http://rest-term.com/archives/3393/\n" }, { "path": "cv/lsd/jsconfig.json", "content": "{\n \"compilerOptions\": {\n \"target\": \"es6\",\n \"module\": \"commonjs\"\n }\n}" }, { "path": "cv/lsd/line-segment-detector/.gitignore", "content": "node_modules\r\ndst\r\n" }, { "path": "cv/lsd/line-segment-detector/README.md", "content": "# Line Segment Detector Module\n\n## description\n\nLine detection using Line Segment Detector ([LSD][IPOL]) algorithm\n\n![lsd](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/lsd_apply.jpg)\n\n\nsee also [blog entry][entry]\n\n## install\n\n```\nnpm install line-segment-detector.js\n```\n\n- Todo npm publish ready: add package.json, unit test\n\n### demo\n[Line Segment Detector Demo](http://rest-term.com/labs/html5/lsd/index.html) [(mirror)](https://secret-nether-5891.herokuapp.com/lsd/)\n\n[![lsd_demo](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/lsd_demo.jpg)](http://rest-term.com/labs/html5/lsd/index.html)\n\n\n## usage in JavaScript (ES2015)\n```js\nimport LSD from './lsd';\n// input 8bit gray scale image\nimport SampleImage from './sample.jpg';\n\nconst image = new Image();\nimage.src = SampleImage;\nimage.onload = () => {\n const width = image.width;\n const height = image.height;\n const canvas = document.createElement('canvas');\n const context = canvas.getContext('2d');\n canvas.width = width;\n canvas.height = height;\n context.drawImage(image, 0, 0, width, height);\n document.getElementById('content').appendChild(canvas);\n const imageData = context.getImageData(0, 0, width, height);\n const detector = new LSD();\n const lines = detector.detect(imageData);\n console.log('lines: ' + lines.length.toString());\n detector.drawSegments(context, lines);\n};\n```\n\n### React Component\n```js\nimport React from 'react';\nimport ReactDOM from 'react-dom';\nimport MuiThemeProvider from 'material-ui/styles/MuiThemeProvider';\nimport LSDComponent from './lsd_component';\nimport SampleImage from './sample.jpg';\n\nReactDOM.render(\n \n \n ,\n document.getElementById('content')\n);\n```\n\n\nlicense\n----------\nCopyright © 2017 wellflat Licensed under the [MIT License][MIT]\n\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[IPOL]: http://www.ipol.im/pub/art/2012/gjmr-lsd/\n[entry]: http://rest-term.com/archives/3393/\n" }, { "path": "cv/lsd/line-segment-detector/jsconfig.json", "content": "{\n \"compilerOptions\": {\n \"target\": \"es6\",\n \"module\": \"commonjs\"\n }\n}" }, { "path": "cv/lsd/line-segment-detector/package.json", "content": "{\n \"name\": \"line-segment-detector.js\",\n \"version\": \"1.0.0\",\n \"description\": \"Line Segment Detector\",\n \"main\": \"./dst/lsd.js\",\n \"types\": \"./dst/lsd.d.ts\",\n \"files\": [\n \"dst\"\n ],\n \"scripts\": {\n \"test\": \"echo \\\"Error: no test specified\\\" && exit 1\",\n \"prod\": \"rimraf dst && webpack -p\",\n \"dev\": \"rimraf dst && webpack -d\",\n \"prepublish\": \"npm run prod\"\n },\n \"keywords\": [],\n \"author\": \"wellflat\",\n \"license\": \"MIT\",\n \"dependencies\": {},\n \"devDependencies\": {\n \"eslint\": \"^4.5.0\",\n \"rimraf\": \"^2.6.1\",\n \"ts-loader\": \"^2.3.3\",\n \"typescript\": \"^2.4.2\",\n \"webpack\": \"^3.5.5\"\n }\n}\n" }, { "path": "cv/lsd/line-segment-detector/src/funcs.ts", "content": "import { Edge } from './types';\n\n/**\n * constants and utility math functions\n */\n\nconst M_3_2_PI = (3 * Math.PI) / 2;\nconst M_2__PI = (2 * Math.PI);\nconst M_LN10 = 2.30258509299404568402;\nconst NOT_DEF = -1024.0;\nconst USED = 1;\nconst NOT_USED = 0;\nconst RELATIVE_ERROR_FACTOR = 100.0;\nconst DEG_TO_RADS = Math.PI / 180;\nconst REFINE_NONE = 0;\nconst REFINE_STD = 1;\nconst REFINE_ADV = 2;\n\n\nconst logGamma = (x: number) => x > 15.0 ? logGammaWindschitl(x) : logGammaLanczos(x);\n\nconst logGammaWindschitl = (x: number) =>\n 0.918938533204673 + (x - 0.5) * Math.log(x) - x\n + 0.5 * x * Math.log(x * Math.sinh(1 / x) + 1 / (810.0 * Math.pow(x, 6.0)));\n\nfunction logGammaLanczos(x: number) {\n const q = [\n 75122.6331530, 80916.6278952, 36308.2951477,\n 8687.24529705, 1168.92649479, 83.8676043424,\n 2.50662827511\n ];\n let a = (x + 0.5) * Math.log(x + 5.5) - (x + 5.5);\n let b = 0;\n for(let n = 0; n < 7; ++n) {\n a -= Math.log(x + n);\n b += q[n] * Math.pow(x, n);\n }\n return a + Math.log(b);\n}\n\nconst distSq = (x1: number, x2: number, y1: number, y2: number) =>\n (x2 - x1) * (x2 - x1) + (y2 - y1) * (y2 - y1);\n\nconst dist = (x1: number, x2: number, y1: number, y2: number) => Math.sqrt(distSq(x1, y1, x2, y2));\n\nfunction angleDiffSigned(a: number, b: number) {\n let diff = a - b;\n const PI = Math.PI;\n while (diff <= -PI) {\n diff += M_2__PI;\n }\n while (diff > PI) {\n diff -= M_2__PI;\n }\n return diff;\n}\n\nfunction angleDiff(a: number, b: number) {\n const value = angleDiffSigned(a, b);\n return value >= 0 ? value : -value;\n}\n\nfunction doubleEqual(a: number, b: number) {\n if (a == b) {\n return true;\n }\n const diff = a - b;\n const absDiff = diff >= 0 ? diff : -diff;\n const aa = a >= 0 ? a : - a;\n const bb = b >= 0 ? b : - b;\n let absMax = (aa > bb) ? aa : bb;\n const MIN_VALUE = Number.MIN_VALUE;\n if (absMax < MIN_VALUE) {\n absMax = MIN_VALUE;\n }\n return (absDiff / absMax) <= (RELATIVE_ERROR_FACTOR * Number.EPSILON);\n}\n\nfunction AsmallerB_XoverY(a: Edge, b: Edge) {\n if (a.p.x == b.p.x) {\n return Number(a.p.y < b.p.y);\n } else {\n return Number(a.p.x < b.p.x);\n }\n}\n\nexport {\n M_3_2_PI, M_2__PI, M_LN10,\n NOT_DEF, USED, NOT_USED, RELATIVE_ERROR_FACTOR, DEG_TO_RADS,\n REFINE_NONE, REFINE_STD, REFINE_ADV,\n logGamma, dist, distSq, angleDiff, doubleEqual, AsmallerB_XoverY\n};\n" }, { "path": "cv/lsd/line-segment-detector/src/lsd.ts", "content": "/**\n * Line Segment Detector (LSD) module\n * @author https://github.com/wellflat\n */\n\nimport * as funcs from './funcs';\nimport { Vec4, Point, CoorList, RegionPoint, Rect, Edge } from './types';\n\n/**\n * Create a LineSegmentDetector object.\n * Specifying scale, number of subdivisions for the image,\n * should the lines be refined and other constants as follows:\n *\n * @param refine How should the lines found be refined?\n * REFINE_NONE - No refinement applied.\n * REFINE_STD - Standard refinement is applied. E.g. breaking arches into smaller line approximations.\n * REFINE_ADV - Advanced refinement. Number of false alarms is calculated,\n * lines are refined through increase of precision, decrement in size, etc.\n * @param scale The scale of the image that will be used to find the lines. Range (0..1].\n * @param sigmaScale Sigma for Gaussian filter is computed as sigma = _sigma_scale/_scale.\n * @param quant Bound to the quantization error on the gradient norm.\n * @param angTh Gradient angle tolerance in degrees.\n * @param logEps Detection threshold: -log10(NFA) > _log_eps\n * @param densityTh Minimal density of aligned region points in rectangle.\n * @param nBins Number of bins in pseudo-ordering of gradient modulus.\n */\nexport default class LSD {\n image?: ImageData;\n private imageData?: Uint8ClampedArray;\n width = 0;\n height = 0;\n list: CoorList[] = [];\n angles?: Float64Array;\n modgrad?: Float64Array;\n used?: Uint8Array;\n\n constructor(\n public refineType = funcs.REFINE_NONE, public scale = 0.8, public sigmaScale = 0.6, public quant = 2.0,\n public angTh = 22.5, public logEps = 0.0, public densityTh = 0.7, public nBins = 1024\n ) { }\n /**\n * Detect lines in the input image.\n * @param {ImageData} image\n * @return {Vec4[]}\n */\n detect(image: ImageData) {\n this.image = image;\n this.width = image.width;\n this.height = image.height;\n const lines = this.lsd();\n return lines;\n }\n /**\n * Draws the line segments on a given image.\n * @param {CanvasRenderingContext2D} context\n * @param {Vec4[]} lines\n * @param {string} color\n */\n drawSegments(context: CanvasRenderingContext2D, lines: Vec4[], color = '#ff0000') {\n context.strokeStyle = color;\n context.lineWidth = 1;\n lines.forEach(v => {\n context.beginPath();\n context.moveTo(v.x1, v.y1);\n context.lineTo(v.x2, v.y2);\n context.stroke();\n context.closePath();\n });\n }\n\n /**\n * for debug\n * @param {CanvasRenderingContext2D} context\n */\n putImageData(context: CanvasRenderingContext2D) {\n let src = this.imageData,\n image = context.createImageData(this.width, this.height),\n dst = image.data,\n len = image.data.length;\n if (!src) throw new Error('imageData required'); // type guard\n for (let i = 0; i < len; i += 4) {\n dst[i] = dst[i + 1] = dst[i + 2] = src[i/4];\n dst[i + 3] = 255;\n }\n context.putImageData(image, 0, 0);\n }\n /**\n * @return {Vec4[]} Return: A vector of Vec4f elements specifying the beginning and ending point of a line.\n * Where Vec4f is (x1, y1, x2, y2), point 1 is the start, point 2 - end.\n * Returned lines are strictly oriented depending on the gradient.\n */\n lsd() {\n if (!this.image || !this.angles) throw new Error('image and angles required'); // type guard\n /** @type {Vec4[]} */\n let lines = [];\n const prec = Math.PI * this.angTh / 180;\n const p = this.angTh / 180;\n const rho = this.quant / Math.sin(prec);\n if (this.scale != 1) {\n const sigma = this.scale < 1 ? this.sigmaScale / this.scale : this.sigmaScale;\n const sprec = 3;\n const h = Math.ceil(sigma * Math.sqrt(2 * sprec * Math.log(10.0)));\n const kSize = 1 + 2 * h;\n const reshaped = this.reshape(this.image);\n this.imageData = this.gaussianBlur(reshaped, kSize, sigma);\n this.computeLevelLineAngles(rho, this.nBins);\n } else {\n this.imageData = this.reshape(this.image);\n this.computeLevelLineAngles(rho, this.nBins);\n }\n\n const LOG_NT = 5 * (Math.log10(this.width) + Math.log10(this.height)) / 2 + Math.log10(11.0);\n const minRegSize = -LOG_NT / Math.log10(p);\n this.used = new Uint8Array(this.imageData.length);\n for (let i = 0, listSize = this.list.length; i < listSize; i++) {\n const point = this.list[i].p;\n if ((this.at(this.used, point) === funcs.NOT_USED) &&\n (this.at(this.angles, point) !== funcs.NOT_DEF)) {\n let regAngle = 0.0;\n let reg: RegionPoint[] = [];\n regAngle = this.regionGrow(this.list[i].p, reg, regAngle, prec);\n if (reg.length < minRegSize) {\n continue;\n }\n let rect = new Rect();\n this.region2Rect(reg, regAngle, prec, p, rect);\n let logNfa = -1;\n if (this.refineType > funcs.REFINE_NONE) {\n if (!this.refine(reg, regAngle, prec, p, rect, this.densityTh)) {\n continue;\n }\n if (this.refineType >= funcs.REFINE_ADV) {\n logNfa = this.improveRect(rect);\n if (logNfa <= this.logEps) {\n continue;\n }\n }\n }\n rect.x1 += 0.5;\n rect.y1 += 0.5;\n rect.x2 += 0.5;\n rect.y2 += 0.5;\n /*\n if (this.scale != 1) {\n rect.x1 /= this.scale;\n rect.y1 /= this.scale;\n rect.x2 /= this.scale;\n rect.y2 /= this.scale;\n rect.width /= this.scale;\n }\n */\n lines.push(new Vec4(rect.x1, rect.y1, rect.x2, rect.y2));\n }\n }\n return lines;\n }\n\n /**\n * @param {number} threshold The minimum value of the angle that is considered defined, otherwise NOTDEF\n * @param {number} nBins The number of bins with which gradients are ordered by, using bucket sort.\n */\n computeLevelLineAngles(threshold: number, nBins: number) {\n const imageData = this.imageData;\n if (!imageData) throw new Error('imageData required'); // type guard\n const width = this.width;\n const height = this.height;\n this.angles = new Float64Array(imageData.length);\n this.modgrad = new Float64Array(imageData.length);\n this.angles = this.setRow(this.angles, height - 1, funcs.NOT_DEF);\n this.angles = this.setCol(this.angles, width - 1, funcs.NOT_DEF);\n let maxGrad = -1.0;\n for (let y = 0; y < height - 1; y++) {\n const step = y * width;\n const nextStep = (y + 1) * width;\n for (let x = 0; x < width - 1; x++) {\n const DA = imageData[x + 1 + nextStep] - imageData[x + step];\n const BC = imageData[x + 1 + step] - imageData[x + nextStep];\n const gx = DA + BC;\n const gy = DA - BC;\n const norm = Math.sqrt((gx * gx + gy * gy) / 4.0);\n this.modgrad[x + step] = norm;\n if (norm <= threshold) {\n this.angles[x + step] = funcs.NOT_DEF;\n } else {\n this.angles[x + step] = Math.atan2(gx, -gy);\n if (norm > maxGrad) {\n maxGrad = norm;\n }\n }\n }\n }\n /** @type {CoorList[]} */\n let rangeS = [];\n rangeS.length = nBins;\n /** @type {CoorList[]} */\n let rangeE = [];\n rangeE.length = nBins;\n let count = 0;\n const binCoef = (maxGrad > 0) ? (nBins - 1) / maxGrad : 0;\n for (let y = 0; y < height - 1; y++) {\n let step = y * width;\n for (let x = 0; x < width - 1; x++) {\n let i = Math.floor(this.modgrad[x + step] * binCoef);\n if (!rangeE[i]) {\n this.list[count] = new CoorList();\n rangeE[i] = rangeS[i] = this.list[count];\n count++;\n } else {\n this.list[count] = new CoorList();\n rangeE[i] = this.list[count];\n rangeE[i].next = this.list[count];\n count++;\n }\n rangeE[i].p = new Point(x, y);\n rangeE[i].next = null;\n }\n }\n let idx = nBins - 1;\n for (; idx > 0 && !rangeS[idx]; idx--) {\n // do nothing.\n }\n let start = rangeS[idx];\n let endIdx = idx;\n if (start) {\n while (idx > 0) {\n idx--;\n if (rangeS[idx]) {\n rangeE[endIdx].next = rangeS[idx];\n rangeE[endIdx] = rangeE[idx];\n endIdx = idx;\n }\n }\n }\n }\n\n regionGrow(s: Point, reg: RegionPoint[], regAngle: number, prec: number) {\n if (!this.used || !this.angles || !this.modgrad) throw new Error('used, angles and modgrad required'); // type guard\n let seed = new RegionPoint();\n seed.x = s.x;\n seed.y = s.y;\n seed.used = this.at(this.used, s);\n regAngle = this.at(this.angles, s);\n seed.angle = regAngle;\n seed.modgrad = this.at(this.modgrad, s);\n seed.used = funcs.USED;\n reg.push(seed);\n let sumdx = Math.cos(regAngle);\n let sumdy = Math.sin(regAngle);\n\n for (let i = 0; i < reg.length; i++) {\n const rpoint = reg[i],\n xxMin = Math.max(rpoint.x - 1, 0),\n xxMax = Math.min(rpoint.x + 1, this.width - 1),\n yyMin = Math.max(rpoint.y - 1, 0),\n yyMax = Math.min(rpoint.y + 1, this.height - 1);\n for (let yy = yyMin; yy <= yyMax; yy++) {\n const step = yy * this.width;\n for (let xx = xxMin; xx <= xxMax; xx++) {\n let isUsed = this.used[xx + step];\n if (isUsed != funcs.USED && this.isAligned(xx, yy, regAngle, prec)) {\n const angle = this.angles[xx + step];\n isUsed = funcs.USED;\n this.used[xx + step] = funcs.USED;\n let regionPoint = new RegionPoint(\n xx, yy, angle, this.modgrad[xx + step], isUsed\n );\n reg.push(regionPoint);\n sumdx += Math.cos(angle);\n sumdy += Math.sin(angle);\n regAngle = Math.atan2(sumdy, sumdx);\n }\n }\n }\n }\n return regAngle;\n }\n\n isAligned(x: number, y: number, theta: number, prec: number) {\n if (x < 0 || y < 0 || x >= this.width || y >= this.height) {\n return false;\n }\n const a = this.angles![x + y * this.width];\n if (a === funcs.NOT_DEF) {\n return false;\n }\n let nTheta = theta - a;\n if (nTheta < 0) {\n nTheta = -nTheta;\n }\n if (nTheta > funcs.M_3_2_PI) {\n nTheta -= funcs.M_2__PI;\n if (nTheta < 0) {\n nTheta = -nTheta;\n }\n }\n return nTheta <= prec;\n }\n\n region2Rect(reg: RegionPoint[], regAngle: number, prec: number, p: number, rect: Rect) {\n let x = 0, y = 0, sum = 0;\n for (let i = 0; i < reg.length; i++) {\n const pnt = reg[i];\n const weight = pnt.modgrad;\n x += pnt.x * weight;\n y += pnt.y * weight;\n sum += weight;\n }\n if (sum <= 0) {\n throw new Error('weighted sum must differ from 0');\n }\n x /= sum;\n y /= sum;\n const theta = this.getTheta(reg, x, y, regAngle, prec);\n const dx = Math.cos(theta);\n const dy = Math.sin(theta);\n let lMin = 0, lMax = 0, wMin = 0, wMax = 0;\n for (let i = 0; i < reg.length; i++) {\n let regdx = reg[i].x - x;\n let regdy = reg[i].y - y;\n let l = regdx * dx + regdy * dy;\n let w = -regdx * dy + regdy * dx;\n if (l > lMax) {\n lMax = l;\n } else if (l < lMin) {\n lMin = l;\n }\n if (w > wMax) {\n wMax = w;\n } else if (w < wMin) {\n wMin = w;\n }\n }\n rect.x1 = x + lMin * dx;\n rect.y1 = y + lMin * dy;\n rect.x2 = x + lMax * dx;\n rect.y2 = y + lMax * dy;\n rect.width = wMax - wMin;\n rect.x = x;\n rect.y = y;\n rect.theta = theta;\n rect.dx = dx;\n rect.dy = dy;\n rect.prec = prec;\n rect.p = p;\n if (rect.width < 1.0) {\n rect.width = 1.0;\n }\n }\n\n getTheta(reg: RegionPoint[], x: number, y: number, regAngle: number, prec: number) {\n let ixx = 0.0,\n iyy = 0.0,\n ixy = 0.0;\n for (let i = 0; i < reg.length; i++) {\n const regx = reg[i].x;\n const regy = reg[i].y;\n const weight = reg[i].modgrad;\n let dx = regx - x;\n let dy = regy - y;\n ixx += dy * dy * weight;\n iyy += dx * dx * weight;\n ixy -= dx * dy * weight;\n }\n let check = (funcs.doubleEqual(ixx, 0) && funcs.doubleEqual(iyy, 0) && funcs.doubleEqual(ixy, 0));\n if (check) {\n throw new Error('check if inertia matrix is null');\n }\n let lambda = 0.5 * (ixx + iyy - Math.sqrt((ixx - iyy) * (ixx - iyy) + 4.0 * ixy * ixy));\n let theta = (Math.abs(ixx) > Math.abs(iyy)) ? Math.atan2(lambda - ixx, ixy) :\n Math.atan2(ixy, lambda - iyy);\n if (funcs.angleDiff(theta, regAngle) > prec) {\n theta += Math.PI;\n }\n return theta;\n }\n\n refine(reg: RegionPoint[], regAngle: number, prec: number, p: number, rect: Rect, densityTh: number) {\n let density = reg.length / (funcs.dist(rect.x1, rect.y1, rect.x2, rect.y2) * rect.width);\n if (density >= densityTh) {\n return true;\n }\n let xc = reg[0].x;\n let yc = reg[0].y;\n const angC = reg[0].angle;\n let sum = 0, sSum = 0, n = 0;\n for (let i = 0; i < reg.length; i++) {\n reg[i].used = funcs.NOT_USED;\n if (funcs.dist(xc, yc, reg[i].x, reg[i].y) < rect.width) {\n const angle = reg[i].angle;\n let angD = funcs.angleDiff(angle, angC);\n sum += angD;\n sSum += angD * angD;\n n++;\n }\n let meanAngle = sum / n;\n let tau = 2.0 * Math.sqrt((sSum - 2.0 * meanAngle * sum) / n + meanAngle * meanAngle);\n this.regionGrow(new Point(reg[0].x, reg[0].y), reg, regAngle, tau);\n if (reg.length < 2) {\n return false;\n }\n this.region2Rect(reg, regAngle, prec, p, rect);\n density = reg.length / (funcs.dist(rect.x1, rect.y1, rect.x2, rect.y2) * rect.width);\n if (density < densityTh) {\n return this.reduceRegionRadius(reg, regAngle, prec, p, rect, density, densityTh);\n } else {\n return true;\n }\n }\n return false; // type guard (unreachable)\n }\n\n reduceRegionRadius(reg: RegionPoint[], regAngle: number, prec: number, p: number, rect: Rect, density: number, densityTh: number) {\n let xc = reg[0].x;\n let yc = reg[0].y;\n let radSq1 = funcs.distSq(xc, yc, rect.x1, rect.y1);\n let radSq2 = funcs.distSq(xc, yc, rect.x2, rect.y2);\n let radSq = radSq1 > radSq2 ? radSq1 : radSq2;\n while (density < densityTh) {\n radSq *= 0.75 * 0.75; // reduce region's radius to 75%\n for (let i = 0; i < reg.length; i++) {\n if (funcs.distSq(xc, yc, reg[i].x, reg[i].y) > radSq) {\n // remove point from the region\n reg[i].used = funcs.NOT_USED;\n const tmp = reg[i];\n reg[i] = reg[reg.length - 1];\n reg[reg.length - 1] = tmp;\n reg.pop();\n --i;\n }\n }\n if (reg.length < 2) {\n return false;\n }\n this.region2Rect(reg, regAngle, prec, p, rect);\n density = reg.length / (funcs.dist(rect.x1, rect.y1, rect.x2, rect.y2) * rect.width);\n }\n return true;\n }\n\n improveRect(rect: Rect) {\n let delta = 0.5;\n let delta2 = delta / 2.0;\n let logNfa = this.rectNfa(rect);\n if (logNfa > this.logEps) {\n return logNfa;\n }\n let r = new Rect();\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n r.p /= 2;\n r.prec = r.p * Math.PI;\n let logNfaNew = this.rectNfa(rect);\n if (logNfaNew > logNfa) {\n logNfa = logNfaNew;\n rect.copy(r);\n }\n }\n if (logNfa > this.logEps) {\n return logNfa;\n }\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n if ((r.width - delta) >= 0.5) {\n r.width -= delta;\n let logNfaNew = this.rectNfa(r);\n if (logNfaNew > logNfa) {\n rect.copy(r);\n logNfa = logNfaNew;\n }\n }\n }\n if (logNfa > this.logEps) {\n return logNfa;\n }\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n if ((r.width - delta) >= 0.5) {\n r.x1 -= -r.dy * delta2;\n r.y1 -= r.dx * delta2;\n r.x2 -= -r.dy * delta2;\n r.y2 -= r.dx * delta2;\n r.width -= delta;\n let logNfaNew = this.rectNfa(r);\n if (logNfaNew > logNfa) {\n rect.copy(r);\n logNfa = logNfaNew;\n }\n }\n }\n if (logNfa > this.logEps) {\n return logNfa;\n }\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n if ((r.width - delta) >= 0.5) {\n r.p /= 2;\n r.prec = r.p * Math.PI;\n let logNfaNew = this.rectNfa(r);\n if (logNfaNew > logNfa) {\n rect.copy(r);\n logNfa = logNfaNew;\n }\n }\n }\n return logNfa;\n }\n\n rectNfa(rect: Rect) {\n let totalPts = 0,\n algPts = 0,\n halfWidth = rect.width / 2.0,\n dyhw = rect.dy * halfWidth,\n dxhw = rect.dx * halfWidth,\n orderedX = [\n new Edge(),\n new Edge(),\n new Edge(),\n new Edge()\n ],\n minY = orderedX[0],\n maxY = orderedX[0];\n orderedX[0].p.x = rect.x1 - dyhw;\n orderedX[0].p.y = rect.y1 + dxhw;\n orderedX[1].p.x = rect.x2 - dyhw;\n orderedX[1].p.y = rect.y2 + dxhw;\n orderedX[2].p.x = rect.x2 + dyhw;\n orderedX[2].p.y = rect.y2 - dxhw;\n orderedX[3].p.x = rect.x1 + dyhw;\n orderedX[3].p.y = rect.y1 - dxhw;\n\n orderedX.sort(funcs.AsmallerB_XoverY);\n\n for (let i = 1; i < 4; i++) {\n if (minY.p.y > orderedX[i].p.y) {\n minY = orderedX[i];\n }\n if (maxY.p.y < orderedX[i].p.y) {\n maxY = orderedX[i];\n }\n }\n minY.taken = true;\n let leftmost = null;\n for (let i = 0; i < 4; i++) {\n if (!orderedX[i].taken) {\n if (!leftmost) {\n leftmost = orderedX[i];\n } else if (leftmost.p.x > orderedX[i].p.x) {\n leftmost = orderedX[i];\n }\n }\n }\n if (!leftmost) throw new Error('leftmost error'); // type guard\n leftmost.taken = true;\n let rightmost = null;\n for (let i = 0; i < 4; i++) {\n if (!orderedX[i].taken) {\n if (!rightmost) {\n rightmost = orderedX[i];\n } else if (rightmost.p.x < orderedX[i].p.x) {\n rightmost = orderedX[i];\n }\n }\n }\n if (!rightmost) throw new Error('rightmost error'); // type guard\n rightmost.taken = true;\n let tailp = null;\n for (let i = 0; i < 4; i++) {\n if (!orderedX[i].taken) {\n if (!tailp) {\n tailp = orderedX[i];\n } else if (tailp.p.x > orderedX[i].p.x) {\n tailp = orderedX[i];\n }\n }\n }\n if (!tailp) throw new Error('tailp error'); // type guard\n tailp.taken = true;\n let flstep = (minY.p.y != leftmost.p.y) ?\n (minY.p.x + leftmost.p.x) / (minY.p.y - leftmost.p.y) : 0;\n let slstep = (leftmost.p.y != tailp.p.x) ?\n (leftmost.p.x = tailp.p.x) / (leftmost.p.y - tailp.p.x) : 0;\n let frstep = (minY.p.y != rightmost.p.y) ?\n (minY.p.x - rightmost.p.x) / (minY.p.y - rightmost.p.y) : 0;\n let srstep = (rightmost.p.y != tailp.p.x) ?\n (rightmost.p.x - tailp.p.x) / (rightmost.p.y - tailp.p.x) : 0;\n let lstep = flstep, rstep = frstep;\n let leftX = minY.p.x, rightX = minY.p.x;\n let minIter = minY.p.y;\n let maxIter = maxY.p.y;\n for (let y = minIter; y <= maxIter; y++) {\n if (y < 0 || y >= this.height) {\n continue;\n }\n for (let x = leftX; x <= rightX; x++) {\n if (x < 0 || x >= this.width) {\n continue;\n }\n totalPts++;\n if (this.isAligned(x, y, rect.theta, rect.prec)) {\n algPts++;\n }\n }\n if (y >= leftmost.p.y) {\n lstep = slstep;\n }\n if (y >= rightmost.p.y) {\n rstep = srstep;\n }\n leftX += lstep;\n rightX += rstep;\n }\n return this.nfa(totalPts, algPts, rect.p);\n }\n\n nfa(n: number, k: number, p: number) {\n const LOG_NT = 5 * (Math.log10(this.width) + Math.log10(this.height)) / 2 + Math.log10(11.0);\n if (n == 0 || k == 0) {\n return -LOG_NT;\n }\n if (n == k) {\n return -LOG_NT - n * Math.log10(p);\n }\n let pTerm = p / (1 - p);\n let log1Term = (n + 1) - funcs.logGamma(k + 1)\n - funcs.logGamma(n - k + 1)\n + k * Math.log(p) + (n - k) * Math.log(1.0 - p);\n let term = Math.exp(log1Term);\n if (funcs.doubleEqual(term, 0)) {\n if (k > n * p) {\n return -log1Term / funcs.M_LN10 - LOG_NT;\n } else {\n return -LOG_NT;\n }\n }\n let binTail = term;\n let tolerance = 0.1;\n for (let i = k + 1; i <= n; i++) {\n let binTerm = (n - i + 1) / i;\n let multTerm = binTerm * pTerm;\n term *= multTerm;\n binTail += term;\n if (binTerm < 1) {\n let err = term * ((1 - Math.pow(multTerm, (n - i + 1))) / (1 - multTerm) - 1);\n if (err < tolerance * Math.abs(-Math.log10(binTail) - LOG_NT) * binTail) {\n break;\n }\n }\n }\n return -Math.log10(binTail) - LOG_NT;\n }\n\n gaussianBlur(imageData: Uint8ClampedArray, kSize: number, sigma: number) {\n let width = this.width,\n height = this.height,\n src = imageData,\n ctx = document.createElement('canvas').getContext('2d'),\n tmp = ctx!.createImageData(width, height),\n dst = null,\n kernel = this.getGaussianKernel(kSize, sigma),\n r = (kSize - 1) / 2;\n const tmp2 = this.reshape(tmp);\n dst = new Uint8ClampedArray(tmp2.length);\n // separate 2d-filter\n for (let y = 0; y < height; y++) {\n let step = y * width;\n for (let x = 0; x < width; x++) {\n let buff = 0;\n let i = x + step;\n let k = 0;\n for (let kx = -r; kx <= r; kx++) {\n let px = x + kx;\n if (px <= 0 || width <= px) {\n px = x;\n }\n let j = px + step;\n buff += src[j] * kernel[k];\n k++;\n }\n tmp2[i] = buff;\n }\n }\n\n for (let x = 0; x < width; x++) {\n for (let y = 0; y < height; y++) {\n let step = y * width;\n let buff = 0;\n let i = x + step;\n let k = 0;\n for (let ky = -r; ky <= r; ky++) {\n let py = y + ky;\n let kStep = ky * width;\n if (py <= 0 || height <= py) {\n py = y;\n kStep = 0;\n }\n let j = i + kStep;\n buff += tmp2[j] * kernel[k];\n k++;\n }\n dst[i] = buff;\n }\n }\n return dst;\n }\n\n getGaussianKernel(kSize: number, sigma: number) {\n // 1d-kernel\n let kernel = [];\n let sigmaX = sigma > 0 ? sigma : ((kSize - 1) * 0.5 - 1) * 0.3 + 0.8;\n let scale2X = -0.5 / (sigmaX * sigmaX);\n let sum = 0.0;\n for (let i = 0; i < kSize; i++) {\n let x = i - (kSize - 1) * 0.5;\n kernel[i] = Math.exp(scale2X * x * x);\n sum += kernel[i];\n }\n sum = 1. / sum;\n for (let i = 0; i < kSize; i++) {\n kernel[i] *= sum;\n }\n return kernel;\n }\n\n reshape(image: ImageData) {\n let src = image.data;\n let reshaped = new Uint8ClampedArray(src.length / 4);\n let len = reshaped.length;\n for (let i = 0; i < len; i++) {\n reshaped[i] = src[i * 4];\n }\n return reshaped;\n }\n\n at(data: Uint8Array|Float64Array, p: Point) {\n return data[p.x + (p.y * this.width)];\n }\n\n row(data: Float64Array, rowIndex: number) {\n let i = rowIndex * this.width;\n return data.subarray(i, i + this.width);\n }\n\n setRow(data: Float64Array, index: number, value: number) {\n let from = index * this.width;\n let to = from + this.width;\n for (let i = from; i < to; i++) {\n data[i] = value;\n }\n return data;\n }\n\n setCol(data: Float64Array, index: number, value: number) {\n let to = this.height * this.width;\n let step = this.width;\n for (let i = index; i < to; i += step) {\n data[i] = value;\n }\n return data;\n }\n}" }, { "path": "cv/lsd/line-segment-detector/src/types.ts", "content": "class Vec4 {\n constructor(public x1 = 0, public y1 = 0, public x2 = 0, public y2 = 0) { }\n}\n\nclass Point {\n constructor(public x = 0.0, public y = 0.0) { }\n}\n\nclass CoorList {\n p: Point;\n next: CoorList | null;\n\n constructor() {\n this.p = new Point();\n }\n}\n\nclass RegionPoint {\n constructor(public x = 0, public y = 0, public angle = 0.0, public modgrad = 0.0, public used?: number) { }\n}\n\nclass Rect {\n constructor(\n public x1 = 0, public y1 = 0, public x2 = 0, public y2 = 0,\n public width = 0, public height = 0, public x = 0, public y = 0, public theta = 0,\n public dx = 0, public dy = 0, public prec = 0, public p = 0\n ) { }\n\n copy(rect: Rect) {\n this.x1 = rect.x1;\n this.y1 = rect.y1;\n this.x2 = rect.x2;\n this.y2 = rect.y2;\n this.width = rect.width;\n this.height = rect.height;\n this.x = rect.x;\n this.y = rect.y;\n this.theta = rect.theta;\n this.dx = rect.dx;\n this.dy = rect.dy;\n this.prec = rect.prec;\n this.p = rect.p;\n }\n}\n\nclass Edge {\n p: Point;\n taken?: boolean;\n\n constructor() {\n this.p = new Point();\n }\n}\n\nexport {\n Vec4, Point, CoorList, RegionPoint, Rect, Edge\n};\n" }, { "path": "cv/lsd/line-segment-detector/tsconfig.json", "content": "{\r\n \"compileOnSave\": true,\r\n \"compilerOptions\": {\r\n \"module\": \"commonjs\",\r\n \"target\": \"es5\",\r\n \"declaration\": true,\r\n \"noImplicitAny\": true,\r\n \"noImplicitThis\": true,\r\n \"noImplicitReturns\": true,\r\n \"noImplicitUseStrict\": true,\r\n \"noUnusedLocals\": true,\r\n \"noUnusedParameters\": true,\r\n \"preserveConstEnums\": true,\r\n \"sourceMap\": true,\r\n \"strictNullChecks\": true,\r\n \"outDir\": \"\",\r\n \"lib\": [\r\n \"dom\",\r\n \"es2015\"\r\n ]\r\n },\r\n \"exclude\": [\r\n \"node_modules\",\r\n \"dst\"\r\n ]\r\n}\r\n" }, { "path": "cv/lsd/line-segment-detector/webpack.config.js", "content": "const webpack = require('webpack');\nconst path = require('path');\n\nmodule.exports = {\n entry: path.join(__dirname, './src/lsd.ts'),\n output: {\n path: path.join(__dirname, './dst'),\n filename: 'lsd.js',\n library: 'lineSegmentDetector',\n libraryTarget: 'umd'\n },\n module: {\n rules: [\n {\n test: /\\.ts$/,\n exclude: /(node_modules|bower_components)/,\n loader: 'ts-loader'\n }\n ]\n },\n resolve: {\n extensions: ['.js', '.ts']\n },\n devtool: 'source-map'\n};\n" }, { "path": "cv/lsd/package.json", "content": "{\n \"name\": \"lsd\",\n \"version\": \"1.0.0\",\n \"description\": \"LSD: Line Segment Detector\",\n \"main\": \"src/app.js\",\n \"scripts\": {\n \"test\": \"echo \\\"Error: no test specified\\\" && exit 1\",\n \"prod\": \"node_modules/.bin/webpack -p\",\n \"dev\": \"node_modules/.bin/webpack -d\",\n \"start\": \"node_modules/.bin/webpack-dev-server --hot --inline\"\n },\n \"keywords\": [],\n \"author\": \"wellflat\",\n \"license\": \"ISC\",\n \"babel\": {\n \"presets\": [\n \"es2015\",\n \"react\"\n ]\n },\n \"dependencies\": {\n \"material-ui\": \"^0.17.4\",\n \"react\": \"^15.4.2\",\n \"react-dom\": \"^15.4.2\",\n \"react-tap-event-plugin\": \"^2.0.1\"\n },\n \"devDependencies\": {\n \"babel\": \"^6.23.0\",\n \"babel-core\": \"^6.21.0\",\n \"babel-loader\": \"^9.1.2\",\n \"babel-preset-es2015\": \"^6.18.0\",\n \"babel-preset-react\": \"^6.16.0\",\n \"babel-preset-react-hmre\": \"^1.1.1\",\n \"babel-preset-stage-0\": \"^6.24.1\",\n \"babel-register\": \"^6.18.0\",\n \"eslint\": \"^4.18.2\",\n \"eslint-plugin-react\": \"^6.9.0\",\n \"file-loader\": \"^0.11.0\",\n \"html-loader\": \"^0.4.4\",\n \"html-webpack-plugin\": \"^5.5.0\",\n \"url-loader\": \"^0.5.7\",\n \"webpack\": \"^5.76.0\",\n \"webpack-dev-server\": \">=4.11.1\"\n }\n}\n" }, { "path": "cv/lsd/src/app.js", "content": "import React from 'react';\nimport ReactDOM from 'react-dom';\nimport injectTapEventPlugin from 'react-tap-event-plugin';\nimport MuiThemeProvider from 'material-ui/styles/MuiThemeProvider';\nimport AppBar from 'material-ui/AppBar';\nimport Paper from 'material-ui/Paper';\nimport { RadioButton, RadioButtonGroup } from 'material-ui/RadioButton';\nimport LSDComponent from './lsd_component';\nimport GardenImage from './gardenterrace_gray.jpg';\nimport MidtownImage from './midtown_gray.jpg';\nimport BuildingImage from './building_gray.jpg';\nimport TokyoImage from './tokyostation_gray.jpg';\n\ninjectTapEventPlugin();\n\nconst style = {\n appBar: {\n margin: '10px 0 10px 0'\n },\n paper: {\n display: 'inline',\n margin: '15px'\n },\n radio: {\n fontSize: '1em',\n }\n};\n\nconst App = props => (\n \n
\n \n \n \n changeImage(e)} style={style.radio}>\n \n \n \n \n \n \n
\n \n);\n \nReactDOM.render(\n ,\n document.getElementById('content')\n);\n\nconst changeImage = e => {\n const target = e.target.value;\n let image = null;\n switch (target) {\n case 'image1':\n image = BuildingImage;\n break;\n case 'image2':\n image = MidtownImage;\n break;\n case 'image3':\n image = GardenImage;\n break;\n case 'image4':\n image = TokyoImage;\n break;\n }\n \n ReactDOM.render(\n ,\n document.getElementById('content')\n );\n}\n\n/*\nconst image = new Image();\nimage.src = BuildingImage;\nimage.onload = () => {\n const width = image.width;\n const height = image.height;\n const canvas = document.createElement('canvas');\n const context = canvas.getContext('2d');\n canvas.width = width;\n canvas.height = height;\n context.drawImage(image, 0, 0, width, height);\n document.getElementById('content').appendChild(canvas);\n const imageData = context.getImageData(0, 0, width, height);\n let scale = 0.8;\n const detector = new LSD(0, scale);\n const lines = detector.detect(imageData);\n console.log('lines: ' + lines.length.toString());\n detector.drawSegments(context, lines);\n};\n*/\n" }, { "path": "cv/lsd/src/funcs.js", "content": "/**\n * constants and utility math functions\n */\n\nconst M_3_2_PI = (3 * Math.PI) / 2;\nconst M_2__PI = (2 * Math.PI);\nconst M_LN10 = 2.30258509299404568402;\nconst NOT_DEF = -1024.0;\nconst USED = 1;\nconst NOT_USED = 0;\nconst RELATIVE_ERROR_FACTOR = 100.0;\nconst DEG_TO_RADS = Math.PI / 180;\nconst REFINE_NONE = 0;\nconst REFINE_STD = 1;\nconst REFINE_ADV = 2;\n\n\nconst logGamma = x => x > 15.0 ? logGammaWindschitl(x) : logGammaLanczos(x);\n\nconst logGammaWindschitl = x =>\n 0.918938533204673 + (x - 0.5) * Math.log(x) - x\n + 0.5 * x * Math.log(x * Math.sinh(1 / x) + 1 / (810.0 * Math.pow(x, 6.0)));\n\nfunction logGammaLanczos(x) {\n const q = [\n 75122.6331530, 80916.6278952, 36308.2951477,\n 8687.24529705, 1168.92649479, 83.8676043424,\n 2.50662827511\n ];\n let a = (x + 0.5) * Math.log(x + 5.5) - (x + 5.5);\n let b = 0;\n for(let n = 0; n < 7; ++n) {\n a -= Math.log(x + n);\n b += q[n] * Math.pow(x, n);\n }\n return a + Math.log(b);\n}\n\nconst distSq = (x1, x2, y1, y2) =>\n (x2 - x1) * (x2 - x1) + (y2 - y1) * (y2 - y1);\n\nconst dist = (x1, x2, y1, y2) => Math.sqrt(distSq(x1, y1, x2, y2));\n\n\n/**\n * @param {number} a\n * @param {number} b\n * @return {number}\n */\nfunction angleDiffSigned(a, b) {\n let diff = a - b;\n const PI = Math.PI;\n while (diff <= -PI) {\n diff += M_2__PI;\n }\n while (diff > PI) {\n diff -= M_2__PI;\n }\n return diff;\n}\n\n/**\n * @param {number} a\n * @param {number} b\n * @return {number}\n */\nfunction angleDiff(a, b) {\n const value = angleDiffSigned(a, b);\n return value >= 0 ? value : -value;\n}\n\n/**\n * @param {number} a\n * @param {number} b\n * @return {number}\n */\nfunction doubleEqual(a, b) {\n if (a == b) {\n return true;\n }\n const diff = a - b;\n const absDiff = diff >= 0 ? diff : -diff;\n const aa = a >= 0 ? a : - a;\n const bb = b >= 0 ? b : - b;\n let absMax = (aa > bb) ? aa : bb;\n const MIN_VALUE = Number.MIN_VALUE;\n if (absMax < MIN_VALUE) {\n absMax = MIN_VALUE;\n }\n return (absDiff / absMax) <= (RELATIVE_ERROR_FACTOR * Number.EPSILON);\n}\n\n/**\n * \n * @param {Edge} a\n * @param {Edge} b\n */\nfunction AsmallerB_XoverY(a, b) {\n if (a.p.x == b.p.x) {\n return a.p.y < b.p.y;\n } else {\n return a.p.x < b.p.x;\n }\n}\n\nexport {\n M_3_2_PI, M_2__PI, M_LN10,\n NOT_DEF, USED, NOT_USED, RELATIVE_ERROR_FACTOR, DEG_TO_RADS,\n REFINE_NONE, REFINE_STD, REFINE_ADV,\n logGamma, dist, distSq, angleDiff, doubleEqual, AsmallerB_XoverY\n};" }, { "path": "cv/lsd/src/index.html", "content": "\n\n \n \n Line Segment Detector Demo\n \n \n
\n
\n
\n \n" }, { "path": "cv/lsd/src/lsd.js", "content": "/**\n * Line Segment Detector (LSD) module\n * @author https://github.com/wellflat\n */\n\nimport * as funcs from './funcs';\nimport { Vec4, Point, CoorList, RegionPoint, Rect, Edge } from './types';\n\n/**\n * Create a LineSegmentDetector object.\n * Specifying scale, number of subdivisions for the image,\n * should the lines be refined and other constants as follows:\n *\n * @param refine How should the lines found be refined?\n * REFINE_NONE - No refinement applied.\n * REFINE_STD - Standard refinement is applied. E.g. breaking arches into smaller line approximations.\n * REFINE_ADV - Advanced refinement. Number of false alarms is calculated,\n * lines are refined through increase of precision, decrement in size, etc.\n * @param scale The scale of the image that will be used to find the lines. Range (0..1].\n * @param sigmaScale Sigma for Gaussian filter is computed as sigma = _sigma_scale/_scale.\n * @param quant Bound to the quantization error on the gradient norm.\n * @param angTh Gradient angle tolerance in degrees.\n * @param logEps Detection threshold: -log10(NFA) > _log_eps\n * @param densityTh Minimal density of aligned region points in rectangle.\n * @param nBins Number of bins in pseudo-ordering of gradient modulus.\n */\nexport default class LSD {\n constructor(\n refineType = funcs.REFINE_NONE, scale = 0.8, sigmaScale = 0.6, quant = 2.0,\n angTh = 22.5, logEps = 0.0, densityTh = 0.7, nBins = 1024\n ) {\n this.refineType = refineType;\n this.scale = scale;\n this.sigmaScale = sigmaScale;\n this.quant = quant;\n this.angTh = angTh;\n this.logEps = logEps;\n this.densityTh = densityTh;\n this.nBins = nBins;\n /** @type {ImageData} */\n this.image = null;\n /** @private @type {Uint8ClampedArray} */\n this.imageData = null;\n /** @type {number} */\n this.width = 0;\n /** @type {number} */\n this.height = 0;\n /** @type {CoorList[]} */\n this.list = [];\n /** @type {Float64Array} */\n this.angles = null;\n /** @type {Float64Array} */\n this.modgrad = null;\n /** @type {Uint8Array} */\n this.used = null; \n }\n /**\n * Detect lines in the input image.\n * @param {ImageData} image\n * @return {Vec4[]}\n */\n detect(image) {\n this.image = image;\n this.width = image.width;\n this.height = image.height;\n const lines = this.lsd();\n return lines;\n }\n /**\n * Draws the line segments on a given image.\n * @param {CanvasRenderingContext2D} context\n * @param {Vec4[]} lines\n * @param {string} color\n */\n drawSegments(context, lines, color = '#ff0000') {\n context.strokeStyle = color;\n context.lineWidth = 1;\n lines.forEach(v => {\n context.beginPath();\n context.moveTo(v.x1, v.y1);\n context.lineTo(v.x2, v.y2);\n context.stroke();\n context.closePath();\n });\n }\n\n /**\n * for debug\n * @param {CanvasRenderingContext2D} context \n */ \n putImageData(context) {\n let src = this.imageData,\n image = context.createImageData(this.width, this.height),\n dst = image.data,\n len = image.data.length;\n for (let i = 0; i < len; i += 4) {\n dst[i] = dst[i + 1] = dst[i + 2] = src[i/4];\n dst[i + 3] = 255;\n }\n context.putImageData(image, 0, 0);\n }\n /**\n * @return {Vec4[]} Return: A vector of Vec4f elements specifying the beginning and ending point of a line.\n * Where Vec4f is (x1, y1, x2, y2), point 1 is the start, point 2 - end.\n * Returned lines are strictly oriented depending on the gradient.\n */\n lsd() {\n /** @type {Vec4[]} */\n let lines = [];\n const prec = Math.PI * this.angTh / 180;\n const p = this.angTh / 180;\n const rho = this.quant / Math.sin(prec);\n if (this.scale != 1) {\n const sigma = this.scale < 1 ? this.sigmaScale / this.scale : this.sigmaScale;\n const sprec = 3;\n const h = Math.ceil(sigma * Math.sqrt(2 * sprec * Math.log(10.0)));\n const kSize = 1 + 2 * h;\n const reshaped = this.reshape(this.image);\n this.imageData = this.gaussianBlur(reshaped, kSize, sigma);\n this.computeLevelLineAngles(rho, this.nBins);\n } else {\n this.imageData = this.reshape(this.image);\n this.computeLevelLineAngles(rho, this.nBins);\n }\n\n const LOG_NT = 5 * (Math.log10(this.width) + Math.log10(this.height)) / 2 + Math.log10(11.0);\n const minRegSize = -LOG_NT / Math.log10(p);\n this.used = new Uint8Array(this.imageData.length);\n for (let i = 0, listSize = this.list.length; i < listSize; i++) {\n const point = this.list[i].p;\n if ((this.at(this.used, point) === funcs.NOT_USED) &&\n (this.at(this.angles, point) !== funcs.NOT_DEF)) {\n let regAngle = 0.0;\n /** @type {RegionPoint[]} */\n let reg = [];\n regAngle = this.regionGrow(this.list[i].p, reg, regAngle, prec);\n if (reg.length < minRegSize) {\n continue;\n }\n let rect = new Rect();\n this.region2Rect(reg, regAngle, prec, p, rect);\n let logNfa = -1;\n if (this.refineType > funcs.REFINE_NONE) {\n if (!this.refine(reg, regAngle, prec, p, rect, this.densityTh)) {\n continue;\n }\n if (this.refineType >= funcs.REFINE_ADV) {\n logNfa = this.improveRect(rect);\n if (logNfa <= this.logEps) {\n continue;\n }\n }\n }\n rect.x1 += 0.5;\n rect.y1 += 0.5;\n rect.x2 += 0.5;\n rect.y2 += 0.5;\n /*\n if (this.scale != 1) {\n rect.x1 /= this.scale;\n rect.y1 /= this.scale;\n rect.x2 /= this.scale;\n rect.y2 /= this.scale;\n rect.width /= this.scale;\n }\n */\n lines.push(new Vec4(rect.x1, rect.y1, rect.x2, rect.y2));\n }\n }\n return lines;\n }\n\n /**\n * @param {number} threshold The minimum value of the angle that is considered defined, otherwise NOTDEF\n * @param {number} nBins The number of bins with which gradients are ordered by, using bucket sort.\n */\n computeLevelLineAngles(threshold, nBins) {\n const imageData = this.imageData;\n const width = this.width;\n const height = this.height;\n this.angles = new Float64Array(imageData.length);\n this.modgrad = new Float64Array(imageData.length);\n this.angles = this.setRow(this.angles, height - 1, funcs.NOT_DEF);\n this.angles = this.setCol(this.angles, width - 1, funcs.NOT_DEF);\n let maxGrad = -1.0;\n for (let y = 0; y < height - 1; y++) {\n const step = y * width;\n const nextStep = (y + 1) * width;\n for (let x = 0; x < width - 1; x++) {\n const DA = imageData[x + 1 + nextStep] - imageData[x + step];\n const BC = imageData[x + 1 + step] - imageData[x + nextStep];\n const gx = DA + BC;\n const gy = DA - BC;\n const norm = Math.sqrt((gx * gx + gy * gy) / 4.0);\n this.modgrad[x + step] = norm;\n if (norm <= threshold) {\n this.angles[x + step] = funcs.NOT_DEF;\n } else {\n this.angles[x + step] = Math.atan2(gx, -gy);\n if (norm > maxGrad) {\n maxGrad = norm;\n }\n }\n }\n }\n /** @type {CoorList[]} */\n let rangeS = [];\n rangeS.length = nBins;\n /** @type {CoorList[]} */\n let rangeE = [];\n rangeE.length = nBins;\n let count = 0;\n const binCoef = (maxGrad > 0) ? (nBins - 1) / maxGrad : 0;\n for (let y = 0; y < height - 1; y++) {\n let step = y * width;\n for (let x = 0; x < width - 1; x++) {\n let i = Math.floor(this.modgrad[x + step] * binCoef);\n if (!rangeE[i]) {\n this.list[count] = new CoorList();\n rangeE[i] = rangeS[i] = this.list[count];\n count++;\n } else {\n this.list[count] = new CoorList();\n rangeE[i] = this.list[count];\n rangeE[i].next = this.list[count];\n count++;\n }\n rangeE[i].p = new Point(x, y);\n rangeE[i].next = null;\n }\n }\n let idx = nBins - 1;\n for (; idx > 0 && !rangeS[idx]; idx--) {\n // do nothing.\n }\n let start = rangeS[idx];\n let endIdx = idx;\n if (start) {\n while (idx > 0) {\n idx--;\n if (rangeS[idx]) {\n rangeE[endIdx].next = rangeS[idx];\n rangeE[endIdx] = rangeE[idx];\n endIdx = idx;\n }\n }\n }\n }\n\n /**\n * @param {Point} s\n * @param {RegionPoint[]} reg\n * @param {number} regAngle\n * @param {number} prec\n */\n regionGrow(s, reg, regAngle, prec) {\n let seed = new RegionPoint();\n seed.x = s.x;\n seed.y = s.y;\n seed.used = this.at(this.used, s);\n regAngle = this.at(this.angles, s);\n seed.angle = regAngle;\n seed.modgrad = this.at(this.modgrad, s);\n seed.used = funcs.USED;\n reg.push(seed);\n let sumdx = Math.cos(regAngle);\n let sumdy = Math.sin(regAngle);\n\n for (let i = 0; i < reg.length; i++) {\n const rpoint = reg[i],\n xxMin = Math.max(rpoint.x - 1, 0),\n xxMax = Math.min(rpoint.x + 1, this.width - 1),\n yyMin = Math.max(rpoint.y - 1, 0),\n yyMax = Math.min(rpoint.y + 1, this.height - 1);\n for (let yy = yyMin; yy <= yyMax; yy++) {\n const step = yy * this.width;\n for (let xx = xxMin; xx <= xxMax; xx++) {\n let isUsed = this.used[xx + step];\n if (isUsed != funcs.USED && this.isAligned(xx, yy, regAngle, prec)) {\n const angle = this.angles[xx + step];\n isUsed = funcs.USED;\n this.used[xx + step] = funcs.USED;\n let regionPoint = new RegionPoint(\n xx, yy, angle, this.modgrad[xx + step], isUsed\n );\n reg.push(regionPoint);\n sumdx += Math.cos(angle);\n sumdy += Math.sin(angle);\n regAngle = Math.atan2(sumdy, sumdx);\n }\n }\n }\n }\n return regAngle;\n }\n\n /**\n * @param {number} x\n * @param {number} y\n * @param {number} theta\n * @param {number} prec\n * @return {boolean}\n */\n isAligned(x, y, theta, prec) {\n if (x < 0 || y < 0 || x >= this.width || y >= this.height) {\n return false;\n }\n const a = this.angles[x + y * this.width];\n if (a === funcs.NOT_DEF) {\n return false;\n }\n let nTheta = theta - a;\n if (nTheta < 0) {\n nTheta = -nTheta;\n }\n if (nTheta > funcs.M_3_2_PI) {\n nTheta -= funcs.M_2__PI;\n if (nTheta < 0) {\n nTheta = -nTheta;\n }\n }\n return nTheta <= prec;\n }\n\n /**\n * @param {RegionPoint[]} reg\n * @param {number} regAngle\n * @param {number} prec\n * @param {number} p\n * @param {Rect} rect\n */\n region2Rect(reg, regAngle, prec, p, rect) {\n let x = 0, y = 0, sum = 0;\n for (let i = 0; i < reg.length; i++) {\n const pnt = reg[i];\n const weight = pnt.modgrad;\n x += pnt.x * weight;\n y += pnt.y * weight;\n sum += weight;\n }\n if (sum <= 0) {\n throw new Error('weighted sum must differ from 0');\n }\n x /= sum;\n y /= sum;\n const theta = this.getTheta(reg, x, y, regAngle, prec);\n const dx = Math.cos(theta);\n const dy = Math.sin(theta);\n let lMin = 0, lMax = 0, wMin = 0, wMax = 0;\n for (let i = 0; i < reg.length; i++) {\n let regdx = reg[i].x - x;\n let regdy = reg[i].y - y;\n let l = regdx * dx + regdy * dy;\n let w = -regdx * dy + regdy * dx;\n if (l > lMax) {\n lMax = l;\n } else if (l < lMin) {\n lMin = l;\n }\n if (w > wMax) {\n wMax = w;\n } else if (w < wMin) {\n wMin = w;\n }\n }\n rect.x1 = x + lMin * dx;\n rect.y1 = y + lMin * dy;\n rect.x2 = x + lMax * dx;\n rect.y2 = y + lMax * dy;\n rect.width = wMax - wMin;\n rect.x = x;\n rect.y = y;\n rect.theta = theta;\n rect.dx = dx;\n rect.dy = dy;\n rect.prec = prec;\n rect.p = p;\n if (rect.width < 1.0) {\n rect.width = 1.0;\n }\n }\n\n /**\n * @param {RegionPoint[]} reg\n * @param {number} x\n * @param {number} y\n * @param {number} regAngle\n * @param {number} prec\n * @return {number}\n */\n getTheta(reg, x, y, regAngle, prec) {\n let ixx = 0.0,\n iyy = 0.0,\n ixy = 0.0;\n for (let i = 0; i < reg.length; i++) {\n const regx = reg[i].x;\n const regy = reg[i].y;\n const weight = reg[i].modgrad;\n let dx = regx - x;\n let dy = regy - y;\n ixx += dy * dy * weight;\n iyy += dx * dx * weight;\n ixy -= dx * dy * weight;\n }\n let check = (funcs.doubleEqual(ixx, 0) && funcs.doubleEqual(iyy, 0) && funcs.doubleEqual(ixy, 0));\n if (check) {\n throw new Error('check if inertia matrix is null');\n }\n let lambda = 0.5 * (ixx + iyy - Math.sqrt((ixx - iyy) * (ixx - iyy) + 4.0 * ixy * ixy));\n let theta = (Math.abs(ixx) > Math.abs(iyy)) ? Math.atan2(lambda - ixx, ixy) :\n Math.atan2(ixy, lambda - iyy);\n if (funcs.angleDiff(theta, regAngle) > prec) {\n theta += Math.PI;\n }\n return theta;\n }\n\n /**\n * @param {RegionPoint[]} reg\n * @param {number} regAngle\n * @param {number} prec\n * @param {number} p\n * @param {Rect} rect\n * @param {number} densityTh\n * @return {boolean}\n */ \n refine(reg, regAngle, prec, p, rect, densityTh) {\n let density = reg.length / (funcs.dist(rect.x1, rect.y1, rect.x2, rect.y2) * rect.width);\n if (density >= densityTh) {\n return true;\n }\n let xc = reg[0].x;\n let yc = reg[0].y;\n const angC = reg[0].angle;\n let sum = 0, sSum = 0, n = 0;\n for (let i = 0; i < reg.length; i++) {\n reg[i].used = funcs.NOT_USED;\n if (funcs.dist(xc, yc, reg[i].x, reg[i].y) < reg.width) {\n const angle = reg[i].angle;\n let angD = funcs.angleDiff(angle, angC);\n sum += angD;\n sSum += angD * angD;\n n++;\n }\n let meanAngle = sum / n;\n let tau = 2.0 * Math.sqrt((sSum - 2.0 * meanAngle * sum) / n + meanAngle * meanAngle);\n this.regionGrow(new Point(reg[0].x, reg[0].y), reg, regAngle, tau);\n if (reg.length < 2) {\n return false;\n }\n this.region2Rect(reg, regAngle, prec, p, rect);\n density = reg.length / (funcs.dist(rect.x1, rect.y1, rect.x2, rect.y2) * rect.width);\n if (density < densityTh) {\n return this.reduceRegionRadius(reg, regAngle, prec, p, rect, density, densityTh);\n } else {\n return true;\n }\n }\n }\n\n /**\n * @param {RegionPoint[]} reg\n * @param {number} regAngle\n * @param {number} prec\n * @param {number} p\n * @param {Rect} rect\n * @param {number} density\n * @param {number} densityTh\n * @return {boolean}\n */ \n reduceRegionRadius(reg, regAngle, prec, p, rect, density, densityTh) {\n let xc = reg[0].x;\n let yc = reg[0].y;\n let radSq1 = funcs.distSq(xc, yc, rect.x1, rect.y1);\n let radSq2 = funcs.distSq(xc, yc, rect.x2, rect.y2);\n let radSq = radSq1 > radSq2 ? radSq1 : radSq2;\n while (density < densityTh) {\n radSq *= 0.75 * 0.75; // reduce region's radius to 75%\n for (let i = 0; i < reg.length; i++) {\n if (funcs.distSq(xc, yc, reg[i].x, reg[i].y) > radSq) {\n // remove point from the region\n reg[i].used = funcs.NOT_USED;\n const tmp = reg[i];\n reg[i] = reg[reg.length - 1];\n reg[reg.length - 1] = tmp;\n reg.pop();\n --i;\n }\n }\n if (reg.length < 2) {\n return false;\n }\n this.region2Rect(reg, regAngle, prec, p, rect);\n density = reg.length / (funcs.dist(rect.x1, rect.y1, rect.x2, rect.y2) * rect.width);\n }\n return true;\n }\n\n /**\n * @param {Rect} rect\n * @return {number}\n */ \n improveRect(rect) {\n let delta = 0.5;\n let delta2 = delta / 2.0;\n let logNfa = this.rectNfa(rect);\n if (logNfa > this.logEps) {\n return logNfa;\n }\n let r = new Rect();\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n r.p /= 2;\n r.prec = r.p * Math.PI;\n let logNfaNew = this.rectNfa(rect);\n if (logNfaNew > logNfa) {\n logNfa = logNfaNew;\n rect.copy(r);\n }\n }\n if (logNfa > this.logEps) {\n return logNfa;\n }\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n if ((r.width - delta) >= 0.5) {\n r.width -= delta;\n let logNfaNew = this.rectNfa(r);\n if (logNfaNew > logNfa) {\n rect.copy(r);\n logNfa = logNfaNew;\n }\n }\n }\n if (logNfa > this.logEps) {\n return logNfa;\n }\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n if ((r.width - delta) >= 0.5) {\n r.x1 -= -r.dy * delta2;\n r.y1 -= r.dx * delta2;\n r.x2 -= -r.dy * delta2;\n r.y2 -= r.dx * delta2;\n r.width -= delta;\n let logNfaNew = this.rectNfa(r);\n if (logNfaNew > logNfa) {\n rect.copy(r);\n logNfa = logNfaNew;\n }\n }\n }\n if (logNfa > this.logEps) {\n return logNfa;\n }\n r.copy(rect);\n for (let n = 0; n < 5; n++) {\n if ((r.width - delta) >= 0.5) {\n r.p /= 2;\n r.prec = r.p * Math.PI;\n let logNfaNew = this.rectNfa(r);\n if (logNfaNew > logNfa) {\n rect.copy(r);\n logNfa = logNfaNew;\n }\n }\n }\n return logNfa;\n }\n\n /**\n * @param {Rect} rect\n * @return {number}\n */ \n rectNfa(rect) {\n let totalPts = 0,\n algPts = 0,\n halfWidth = rect.width / 2.0,\n dyhw = rect.dy * halfWidth,\n dxhw = rect.dx * halfWidth,\n orderedX = [\n new Edge(),\n new Edge(),\n new Edge(),\n new Edge()\n ],\n minY = orderedX[0],\n maxY = orderedX[0];\n orderedX[0].p.x = rect.x1 - dyhw;\n orderedX[0].p.y = rect.y1 + dxhw;\n orderedX[1].p.x = rect.x2 - dyhw;\n orderedX[1].p.y = rect.y2 + dxhw;\n orderedX[2].p.x = rect.x2 + dyhw;\n orderedX[2].p.y = rect.y2 - dxhw;\n orderedX[3].p.x = rect.x1 + dyhw;\n orderedX[3].p.y = rect.y1 - dxhw;\n\n orderedX.sort(funcs.AsmallerB_XoverY);\n \n for (let i = 1; i < 4; i++) {\n if (minY.p.y > orderedX[i].p.y) {\n minY = orderedX[i];\n }\n if (maxY.p.y < orderedX[i].p.y) {\n maxY = orderedX[i];\n }\n }\n minY.taken = true;\n let leftmost = null;\n for (let i = 0; i < 4; i++) {\n if (!orderedX[i].taken) {\n if (!leftmost) {\n leftmost = orderedX[i];\n } else if (leftmost.p.x > orderedX[i].p.x) {\n leftmost = orderedX[i];\n }\n }\n }\n leftmost.taken = true;\n let rightmost = null;\n for (let i = 0; i < 4; i++) {\n if (!orderedX[i].taken) {\n if (!rightmost) {\n rightmost = orderedX[i];\n } else if (rightmost.p.x < orderedX[i].p.x) {\n rightmost = orderedX[i];\n }\n }\n }\n rightmost.taken = true;\n let tailp = null;\n for (let i = 0; i < 4; i++) {\n if (!orderedX[i].taken) {\n if (!tailp) {\n tailp = orderedX[i];\n } else if (tailp.p.x > orderedX[i].p.x) {\n tailp = orderedX[i];\n }\n }\n }\n tailp.taken = true;\n let flstep = (minY.p.y != leftmost.p.y) ?\n (minY.p.x + leftmost.p.x) / (minY.p.y - leftmost.p.y) : 0;\n let slstep = (leftmost.p.y != tailp.p.x) ?\n (leftmost.p.x = tailp.p.x) / (leftmost.p.y - tailp.p.x) : 0;\n let frstep = (minY.p.y != rightmost.p.y) ?\n (minY.p.x - rightmost.p.x) / (minY.p.y - rightmost.p.y) : 0;\n let srstep = (rightmost.p.y != tailp.p.x) ?\n (rightmost.p.x - tailp.p.x) / (rightmost.p.y - tailp.p.x) : 0;\n let lstep = flstep, rstep = frstep;\n let leftX = minY.p.x, rightX = minY.p.x;\n let minIter = minY.p.y;\n let maxIter = maxY.p.y;\n for (let y = minIter; y <= maxIter; y++) {\n if (y < 0 || y >= this.height) {\n continue;\n }\n for (let x = leftX; x <= rightX; x++) {\n if (x < 0 || x >= this.width) {\n continue;\n }\n totalPts++;\n if (this.isAligned(x, y, rect.theta, rect.prec)) {\n algPts++;\n }\n }\n if (y >= leftmost.p.y) {\n lstep = slstep;\n }\n if (y >= rightmost.p.y) {\n rstep = srstep;\n }\n leftX += lstep;\n rightX += rstep;\n }\n return this.nfa(totalPts, algPts, rect.p);\n }\n\n /**\n * @param {number} n\n * @param {number} k\n * @param {number} p\n */ \n nfa(n, k, p) {\n const LOG_NT = 5 * (Math.log10(this.width) + Math.log10(this.height)) / 2 + Math.log10(11.0);\n if (n == 0 || k == 0) {\n return -LOG_NT;\n }\n if (n == k) {\n return -LOG_NT - n * Math.log10(p);\n }\n let pTerm = p / (1 - p);\n let log1Term = (n + 1) - funcs.logGamma(k + 1)\n - funcs.logGamma(n - k + 1)\n + k * Math.log(p) + (n - k) * Math.log(1.0 - p);\n let term = Math.exp(log1Term);\n if (funcs.doubleEqual(term, 0)) {\n if (k > n * p) {\n return -log1Term / funcs.M_LN10 - LOG_NT;\n } else {\n return -LOG_NT;\n }\n }\n let binTail = term;\n let tolerance = 0.1;\n for (let i = k + 1; i <= n; i++) {\n let binTerm = (n - i + 1) / i;\n let multTerm = binTerm * pTerm;\n term *= multTerm;\n binTail += term;\n if (binTerm < 1) {\n let err = term * ((1 - Math.pow(multTerm, (n - i + 1))) / (1 - multTerm) - 1);\n if (err < tolerance * Math.abs(-Math.log10(binTail) - LOG_NT) * binTail) {\n break;\n }\n }\n }\n return -Math.log10(binTail) - LOG_NT;\n }\n\n /**\n * @param {Uint8ClampedArray} imageData\n * @param {number} kSize\n * @param {number} sigma\n * @return {Uint8ClampedArray}\n */ \n gaussianBlur(imageData, kSize, sigma) {\n let width = this.width,\n height = this.height,\n src = imageData,\n ctx = document.createElement('canvas').getContext('2d'),\n tmp = ctx.createImageData(width, height),\n dst = null,\n kernel = this.getGaussianKernel(kSize, sigma),\n r = (kSize - 1) / 2;\n tmp = this.reshape(tmp);\n dst = new Uint8ClampedArray(tmp.length);\n // separate 2d-filter\n for (let y = 0; y < height; y++) {\n let step = y * width;\n for (let x = 0; x < width; x++) {\n let buff = 0;\n let i = x + step;\n let k = 0;\n for (let kx = -r; kx <= r; kx++) {\n let px = x + kx;\n if (px <= 0 || width <= px) {\n px = x;\n }\n let j = px + step;\n buff += src[j] * kernel[k];\n k++;\n }\n tmp[i] = buff;\n }\n }\n \n for (let x = 0; x < width; x++) {\n for (let y = 0; y < height; y++) {\n let step = y * width;\n let buff = 0;\n let i = x + step;\n let k = 0;\n for (let ky = -r; ky <= r; ky++) {\n let py = y + ky;\n let kStep = ky * width;\n if (py <= 0 || height <= py) {\n py = y;\n kStep = 0;\n }\n let j = i + kStep;\n buff += tmp[j] * kernel[k];\n k++;\n }\n dst[i] = buff;\n }\n }\n return dst;\n }\n\n /**\n * @param {number} kSize\n * @param {number} sigma\n * @return {number[]}\n */\n getGaussianKernel(kSize, sigma) {\n // 1d-kernel\n let kernel = [];\n let sigmaX = sigma > 0 ? sigma : ((kSize - 1) * 0.5 - 1) * 0.3 + 0.8;\n let scale2X = -0.5 / (sigmaX * sigmaX);\n let sum = 0.0;\n for (let i = 0; i < kSize; i++) {\n let x = i - (kSize - 1) * 0.5;\n kernel[i] = Math.exp(scale2X * x * x);\n sum += kernel[i];\n }\n sum = 1. / sum;\n for (let i = 0; i < kSize; i++) {\n kernel[i] *= sum;\n }\n return kernel;\n }\n /**\n * @param {ImageData} image\n * @return {Uint8ClampedArray}\n */\n reshape(image) {\n let src = image.data;\n let reshaped = new Uint8ClampedArray(src.length / 4);\n let len = reshaped.length;\n for (let i = 0; i < len; i++) {\n reshaped[i] = src[i * 4];\n }\n return reshaped;\n }\n\n /**\n * @param {Uint8Array|Float64Array} data\n * @param {Point} p \n */\n at(data, p) {\n return data[p.x + (p.y * this.width)];\n }\n\n /**\n * @param {Float64Array} data\n * @param {number} rowIndex\n */\n row(data, rowIndex) {\n let i = rowIndex * this.width; \n return data.subarray(i, i + this.width);\n }\n\n /**\n * @param {Float64Array} data\n * @param {number} index\n * @param {number} value\n * @return {null}\n */\n setRow(data, index, value) {\n let from = index * this.width;\n let to = from + this.width;\n for (let i = from; i < to; i++) {\n data[i] = value;\n }\n return data;\n }\n\n /**\n * @param {Float64Array} data\n * @param {number} index\n * @param {number} value\n * @return {null}\n */\n setCol(data, index, value) {\n let to = this.height * this.width;\n let step = this.width;\n for (let i = index; i < to; i += step) {\n data[i] = value;\n }\n return data;\n }\n}" }, { "path": "cv/lsd/src/lsd_component.js", "content": "import React from 'react';\nimport RaisedButton from 'material-ui/RaisedButton';\nimport LSD from './lsd';\n\nexport default class LSDComponent extends React.Component {\n constructor(props) {\n super(props);\n this.onLoad = this.onLoad.bind(this);\n }\n componentDidMount() {\n this.image = new Image();\n this.image.onload = this.onLoad;\n this.image.src = this.props.src;\n this.canvas = this.refs.canvas;\n this.width = 0;\n this.height = 0;\n }\n componentDidUpdate() {\n this.image.src = this.props.src;\n }\n apply(e) {\n e.preventDefault();\n const ctx = this.canvas.getContext('2d'),\n imageData = ctx.getImageData(0, 0, this.width, this.height),\n scale = 0.8,\n detector = new LSD(0, scale),\n lines = detector.detect(imageData);\n console.log('lines: ' + lines.length.toString());\n detector.drawSegments(ctx, lines);\n }\n reset(e) {\n e.preventDefault();\n const ctx = this.canvas.getContext('2d');\n ctx.drawImage(this.image, 0, 0, this.width, this.height);\n }\n onLoad() {\n const ctx = this.canvas.getContext('2d');\n this.width = this.image.width;\n this.height = this.image.height;\n this.canvas.width = this.width;\n this.canvas.height = this.height;\n ctx.drawImage(this.image, 0, 0, this.width, this.height);\n }\n render() {\n return (\n
\n
\n \n
\n
\n this.apply(e)} />\n this.reset(e)} />\n
\n
\n )\n }\n}" }, { "path": "cv/lsd/src/types.js", "content": "class Vec4 {\n constructor(x1 = 0, y1 = 0, x2 = 0, y2 = 0) {\n this.x1 = x1;\n this.y1 = y1;\n this.x2 = x2;\n this.y2 = y2;\n }\n}\n\nclass Point {\n constructor(x = 0.0, y = 0.0) {\n this.x = x;\n this.y = y;\n }\n}\n\nclass CoorList {\n constructor() {\n this.p = new Point();\n /** @type {CoorList} */\n this.next = null;\n }\n}\n\nclass RegionPoint {\n constructor(x = 0, y = 0, angle = 0.0, modgrad = 0.0, used = null) {\n this.x = x;\n this.y = y;\n this.angle = angle;\n this.modgrad = modgrad;\n /** @type {number} */\n this.used = used;\n }\n}\n\nclass Rect {\n constructor(\n x1 = 0, y1 = 0, x2 = 0, y2 = 0,\n width = 0, height = 0, x = 0, y = 0, theta = 0,\n dx = 0, dy = 0, prec = 0, p = 0\n ) {\n this.x1 = x1;\n this.y1 = y1;\n this.x2 = x2;\n this.y2 = y2;\n this.width = width;\n this.height = height;\n this.x = x;\n this.y = y;\n this.theta = theta;\n this.dx = dx;\n this.dy = dy;\n this.prec = prec;\n this.p = p;\n }\n /**\n * @param {Rect} rect\n */\n copy(rect) {\n this.x1 = rect.x1;\n this.y1 = rect.y1;\n this.x2 = rect.x2;\n this.y2 = rect.y2;\n this.width = rect.width;\n this.height = rect.height;\n this.x = rect.x;\n this.y = rect.y;\n this.theta = rect.theta;\n this.dx = rect.dx;\n this.dy = rect.dy;\n this.prec = rect.prec;\n this.p = rect.p;\n }\n}\n\nclass Edge {\n constructor() {\n this.p = new Point();\n this.taken = null;\n }\n}\n\nexport {\n Vec4, Point, CoorList, RegionPoint, Rect, Edge\n};" }, { "path": "cv/lsd/webpack.config.js", "content": "const HtmlWebpackPlugin = require('html-webpack-plugin');\nconst webpack = require('webpack');\n\nmodule.exports = {\n entry: __dirname + '/src/app.js',\n output: {\n path: __dirname + '/dst',\n filename: 'bundle.js',\n },\n module: {\n loaders: [\n {\n test: /\\.js[x]?$/,\n exclude: /(node_modules|bower_components)/,\n loader: 'babel-loader',\n query: {\n presets: ['es2015', 'react'],//, 'react-hmre']\n }\n },\n {\n test: /\\.html$/,\n loader: 'html-loader'\n },\n {\n test: /\\.(jpg|png)$/,\n loader: 'url-loader'\n }\n ]\n },\n resolve: {\n extensions: ['.js', '.jsx']\n },\n devServer: {\n contentBase: '/dst',\n port: 8888\n },\n devtool: 'source-map',\n plugins: [\n new HtmlWebpackPlugin({\n title: \"React Sample\",\n filename: 'index.html',\n template: __dirname + '/src/index.html'\n }) \n ]\n};" }, { "path": "cv/ort_web/HelloONNX.vue", "content": "\n\n" }, { "path": "cv/ort_web/README.md", "content": "## Image Classification using ONNX Runtime for Web (ORT Web) \n\nthis example is the [CIFAR-10](https://www.cs.toronto.edu/~kriz/cifar.html) classification build by [Vue.js](https://vuejs.org/index.html) component\n\nsee also: [ONNX Runtime Web](https://github.com/microsoft/onnxruntime/tree/master/js/web#readme)" }, { "path": "cv/pixel_clustering/README.md", "content": "# Pixel Clustering Module\n\n## description\n\nPixel clustering using k-means++ algorithm\n\nsee also [blog entry][entry]\n\n### sample\n[![pixel_clustering](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/pixel_clustering.jpg)](http://rest-term.com/labs/html5/pixelclustering.html)\n\n(Tests: IE10, Firefox23.0, Chrome28.0, Safari6.0)\n\n## usage\n\n```js\nvar ctx = document.querySelector('#Canvas').getContext('2d');\n// source image, callback function\nPixelCluster.load(srcImgName, function() {\n // performs k-means on a set of observation vectors forming k clusters\n var division = 100;\n var k = 4;\n var method = PixelCluster.KMEANS_PP; // KMEANS_PP or KMEANS_RANDOM\n PixelCluster.perform(division, k, method, function(result) {\n // render the result to canvas element\n PixelCluster.render(ctx, division, result);\n });\n});\n```\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[entry]: http://rest-term.com/archives/3073/\n" }, { "path": "cv/pixel_clustering/kmeans.js", "content": "/**\n * provides routines for k-means clustering, generating code books,\n * and quantizing vectors by comparing them with centroids.\n * (specified CanvasPixelArray [[R,G,B],[R,G,B],...])\n */\n\n/* performs k-means on a set of observation vectors forming k clusters */\nfunction kmeans(samples, ncluster, method) {\n var centroids = [],\n previous = [],\n clusters = [],\n code = [],\n len = samples.length,\n eps = 1.0e-8,\n maxIter = 1000,\n iter = 0,\n distance = function(a, b) {\n var dr = a[0] - b[0],\n dg = a[1] - b[1],\n db = a[2] - b[2];\n return dr*dr + dg*dg + db*db;\n },\n nearest = function(sample, centroids) {\n var minIndex = 0,\n minDistance = Number.MAX_VALUE,\n clusterCount = centroids.length;\n for(var k=0; k d) {\n minDistance = d;\n minIndex = k;\n }\n }\n return [minIndex, minDistance];\n },\n initialize = function(samples, ncluster, method, centroids, clusters) {\n var len = samples.length;\n if(method === 'kmeans_pp') {\n // kmeans++\n var d = [], sumDistance = 0.0, label, k, i,\n r = Math.floor(Math.random()*len);\n centroids[0] = [];\n centroids[0][0] = samples[r][0];\n centroids[0][1] = samples[r][1];\n centroids[0][2] = samples[r][2];\n previous[0] = [0.0, 0.0, 0.0];\n clusters[0] = [];\n for(k=1; k 0) continue;\n centroids[k] = [];\n centroids[k][0] = samples[i][0];\n centroids[k][1] = samples[i][1];\n centroids[k][2] = samples[i][2];\n break;\n }\n previous[k] = [0.0, 0.0, 0.0];\n clusters[k] = [];\n }\n for(i=0; i d) {\n minDistance = d;\n minIndex = k;\n }\n }\n return [minIndex, minDistance];\n },\n initialize = function(samples, ncluster, method, centroids, clusters) {\n var len = samples.length;\n if(method === 'kmeans_pp') {\n // kmeans++\n var d = [], sumDistance = 0.0, label, k, i,\n r = Math.floor(Math.random()*len);\n centroids[0] = [];\n centroids[0][0] = samples[r][0];\n centroids[0][1] = samples[r][1];\n centroids[0][2] = samples[r][2];\n previous[0] = [0.0, 0.0, 0.0];\n clusters[0] = [];\n for(k=1; k 0) continue;\n centroids[k] = [];\n centroids[k][0] = samples[i][0];\n centroids[k][1] = samples[i][1];\n centroids[k][2] = samples[i][2];\n break;\n }\n previous[k] = [0.0, 0.0, 0.0];\n clusters[k] = [];\n }\n for(i=0; i threashold) { // on the mask\n sumf = [0.0, 0.0, 0.0];\n sumfstar = [0.0, 0.0, 0.0];\n sumvq = [0.0, 0.0, 0.0];\n edge = false;\n for(var n=0; n<4; n++) {\n if(maskData[naddr[n]] <= threashold) {\n edge = true;\n break;\n }\n }\n if(!edge) {\n if(y + offsetY >= 0 && x + offsetX >= 0 &&\n y + offsetY < h && x + offsetX < w) {\n for(n=0; n<4; n++) {\n for(var c=0; c<3; c++) {\n sumf[c] += blendData[naddr[n] + m + c];\n sumvq[c] += srcData[l + c] - srcData[naddr[n] + c];\n }\n }\n }\n } else {\n if(y + offsetY >= 0 && x + offsetX >= 0 &&\n y + offsetY < h && x + offsetX < w) {\n fp[0] = dstData[l + m];\n fp[1] = dstData[l + m + 1];\n fp[2] = dstData[l + m + 2];\n gp[0] = srcData[l];\n gp[1] = srcData[l + 1];\n gp[2] = srcData[l + 2];\n for(n=0; n<4; n++) {\n for(c=0; c<3; c++) {\n fq[c] = dstData[naddr[n] + m + c];\n gq[c] = srcData[naddr[n] + c];\n sumfstar[c] += fq[c];\n subf[c] = fp[c] - fq[c];\n subf[c] = subf[c] > 0 ? subf[c] : -subf[c];\n subg[c] = gp[c] - gq[c];\n subg[c] = subg[c] > 0 ? subg[c] : -subg[c];\n if(subf[c] > subg[c]) {\n sumvq[c] += subf[c];\n } else {\n sumvq[c] += subg[c];\n }\n }\n }\n }\n }\n for(c=0; c<3; c++) {\n fp[c] = (sumf[c] + sumfstar[c] + sumvq[c])*0.25; // division 4\n error = Math.floor(fp[c] - blendData[l + m + c]);\n error = error > 0 ? error : -error;\n if(terminate[c] && error > EPS*(1 + (fp[c] > 0 ? fp[c] : -fp[c]))) {\n terminate[c] = false;\n }\n blendData[l + m + c] = fp[c];\n }\n } // end mask\n } // end x loop\n } // end y loop\n if(terminate[0] && terminate[1] && terminate[2]) break;\n } // end iteration\n return data[3];\n }\n };\n // aliases (public APIs)\n Poisson.load = core.load;\n Poisson.reset = core.reset;\n Poisson.blend = core.blend;\n}).call(this);\n" }, { "path": "cv/stereo_matching/README.md", "content": "# Stereo Matching Module\n\n## description\n\ncomputes the disparity map using naive block matching\n\nsee also [blog entry][entry]\n\n### sample\n[![stereo_matching](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/stereo_matching.jpg)](http://rest-term.com/labs/html5/stereo.html)\n\n(Tests: IE10, Firefox7.0, Chrome14.0, Opera11.51)\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[entry]: http://rest-term.com/archives/2940/\n" }, { "path": "cv/stereo_matching/stereo-core.js", "content": "/**\n * computes the disparity for the rectified stereo pair,\n * naive implementation without using dynamic programming.\n */\nfunction findStereoCorrespondence(pair, state) {\n var w = pair[0].width,\n h = pair[0].height,\n data = [pair[0].data, pair[1].data],\n disparity = [],\n step = w << 2,\n winSize = state.SADWindowSize,\n threshold = state.textureThreshold,\n minDisparity = state.minDisparity,\n maxDisparity = state.numberOfDisparities - minDisparity,\n maxDiff = 255 * winSize * winSize,\n localMin = maxDiff,\n sad = 0,\n diff, cur, ptr;\n for(var y=0, cur=step; y threshold || y + ly > h) break;\n for(var lx=0, j=ly*step; lx w) break;\n diff = data[0][i + j] - data[1][ptr + j];\n if(isNaN(diff)) diff = 0;\n sad += (diff < 0 ? -diff : diff);\n }\n }\n if(localMin > sad) {\n localMin = sad;\n disparity[i] = disparity[i + 1] = disparity[i + 2] = d << 4;\n disparity[i + 3] = 255;\n }\n sad = 0;\n }\n localMin = maxDiff;\n }\n }\n return {'width':w, 'height':h, 'disparity':disparity};\n}\n\n/* message handler */\nself.addEventListener('message', function(e) {\n var message = e.data,\n data = message.pair, // CanvasPixelArray pair\n state = message.state;\n postMessage(findStereoCorrespondence(data, state));\n}, false);\nself.addEventListener('error', function(e) {\n postMessage(e);\n}, false);\n" }, { "path": "cv/stereo_matching/stereo.js", "content": "/**\n * Stereo Matching module using Web Workers\n */\n(function() {\n var Stereo; // top-level namespace\n var _root = this; // reference to 'window' or 'global'\n\n if(typeof exports !== 'undefined') {\n Stereo = exports; // for CommonJS\n } else {\n Stereo = _root.Stereo = {};\n }\n\n var version = {\n release: '0.1.0',\n date: '2011-01'\n };\n Stereo.toString = function() {\n return \"version \" + version.release + \", released \" + version.date;\n };\n\n // core operations\n var ctx = null,\n left = new Image(),\n right = new Image(),\n data = [null, null],\n state = {},\n worker = new Worker('lib/js/stereo-core.js');\n var core = {\n /* load stereo pair images */\n _load : function(leftSource, rightSource) {\n ctx = document.createElement('canvas').getContext('2d');\n left.src = leftSource;\n right.src = rightSource;\n // load complete handler\n left.addEventListener('load', function() {\n ctx.canvas.width = left.width;\n ctx.canvas.height = left.height;\n ctx.drawImage(left, 0, 0);\n data[0] = ctx.getImageData(0, 0, left.width, left.height);\n }, false);\n right.addEventListener('load', function() {\n ctx.canvas.width = right.width;\n ctx.canvas.height = right.height;\n ctx.drawImage(right, 0, 0);\n data[1] = ctx.getImageData(0, 0, right.width, right.height);\n }, false);\n // load error handler\n left.addEventListener('error', function() {\n alert('can\\'t load left image');\n }, false);\n right.addEventListener('error', function() {\n alert('can\\'t load right image');\n }, false);\n },\n /* draw disparity map on the canvas */\n _draw : function(data, context) {\n var img = context.getImageData(0, 0, data.width, data.height),\n pixels = img.data,\n disparity = data.disparity,\n len = disparity.length;\n for(var i=0; i max ? this.bins[ch][i] : max;\n }\n }\n return max;\n },\n normalize : function(factor) {\n for(var ch in this.bins) {\n for(var i=0; i<256; i++) {\n this.bins[ch][i] /= factor;\n }\n }\n },\n backProject : function() {\n // TODO\n },\n compare : function(hist, method) {\n // TODO\n switch(method) {\n case 'correl':\n break;\n case 'chisqr':\n break;\n case 'intersect':\n break;\n case 'bhattacharyya':\n break;\n default:\n break;\n }\n },\n convertColor : function(type) {\n // TODO\n switch(type) {\n case 'rgb':\n break;\n case 'gray':\n break;\n case 'hsv':\n break;\n default:\n break;\n }\n }\n});\n\nself.addEventListener('message', function(e) {\n var hist = new Histogram(e.data);\n hist.calculate();\n hist.bins.max = hist.getMax();\n postMessage(hist.bins);\n}, false);\n\nself.addEventListener('error', function(e) {\n postMessage(e);\n}, false);" }, { "path": "draw/README.md", "content": "# 3D Shapes Drawing\n\n## description\n\nDrawing 3D shapes (pseudo 3D)\n\n[![3d_drawing](https://raw.github.com/wiki/wellflat/imageprocessing-labs/images/3dshape_drawing.png)](http://rest-term.com/labs/html5/)\n\n### demo\n[Heart Surface][Heart]\n\n[Mobius Strip][Mobius]\n\n[Klein Bottle][Klein]\n\n(Tests: IE10, Firefox8.0, Safari5.0, Chrome16.0, Opera11.60)\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[Heart]: http://rest-term.com/labs/html5/heart.html\n[Mobius]: http://rest-term.com/labs/html5/mobius.html\n[Klein]: http://rest-term.com/labs/html5/klein.html\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n[entry]: http://rest-term.com/archives/2986/\n" }, { "path": "draw/heart.js", "content": "/**\n * Heart Surface Maker\n */\nvar Heart = (function() {\n var _context = null,\n _width = null,\n _height = null,\n _scale = 25,\n _fov = 250,\n _points = [],\n _numPoints = 0,\n _isPlay = false,\n _timerID = null,\n // public methods\n _init = function(context, n) {\n _context = context;\n _width = context.canvas.width;\n _height = context.canvas.height;\n _points = [];\n _setStyle(_context.createLinearGradient(0, 0, 0, 300));\n _setPoints(n);\n },\n _play = function(t, angle) {\n if(!_isPlay) {\n _timerID = setInterval(function() {\n _render(angle)\n }, t);\n _isPlay = true;\n }\n },\n _stop = function() {\n if(_isPlay) {\n clearInterval(_timerID);\n _isPlay = false;\n }\n },\n // private methods\n _setStyle = function(grad) {\n grad.addColorStop(0, 'rgb(255, 255, 255)');\n grad.addColorStop(0.7, 'rgb(196, 196, 196)');\n grad.addColorStop(0.8, 'rgb(160, 160, 160)');\n grad.addColorStop(0.9, 'rgb(128, 128, 128)');\n grad.addColorStop(1, 'rgb(96, 96, 96)');\n _context.fillStyle = grad;\n _context.strokeStyle = 'rgb(255, 0, 160)';\n _context.fillRect(0, 0, _width, _height);\n },\n _setPoints = function(num) {\n var r, theta, z, n = 0, point;\n for(var i=-Math.PI*num; i<=Math.PI*num; i++) {\n for(var j=-num; j<=num; j++) {\n theta = i/num; z = j/num;\n r = 4*Math.sqrt(1 - z*z)*Math.pow(Math.sin(Math.abs(theta)), Math.abs(theta));\n point = [_scale*r*Math.sin(theta),\n _scale*r*Math.cos(theta),\n _scale*z];\n _points.push(point);\n n++;\n }\n }\n _numPoints = n;\n },\n _render = function(angle) {\n _context.fillRect(0, 0, _width, _height);\n for(var i=0; i<_numPoints; i++)\t{\n _rotateY(_points[i], angle);\n _draw(_points[i]);\n }\n },\n _draw = function(point3d) {\n var scale = _fov/(_fov + point3d[2]),\n x = (point3d[0]*scale) + (_width >> 1),\n y = (point3d[1]*scale) + (_height >> 1);\n _context.lineWidth= scale << 1;\n _context.beginPath();\n _context.moveTo(x, y);\n _context.lineTo(x + scale, y);\n _context.stroke();\n },\n _rotateY = function(point3d, angle) {\n var x = point3d[0],\n z = point3d[2],\n c = Math.cos(angle),\n s = Math.sin(angle),\n tz = z,\n tx = x;\n x = tx*c + tz*s;\n z = tz*c - tx*s;\n point3d[0] = x;\n point3d[2] = z;\n };\n // public APIs\n return {\n init: _init,\n play: _play,\n stop: _stop,\n };\n})();" }, { "path": "draw/klein.js", "content": "/**\n * Klein's Bottle Maker\n */\nvar KleinBottle = (function() {\n var _context = null,\n _width = null,\n _height = null,\n _scale = 6,\n _fov = 250,\n _points = [],\n _numPoints = 0,\n _isPlay = false,\n _timerID = null,\n // public methods\n _init = function(context, n) {\n _context = context;\n _width = context.canvas.width;\n _height = context.canvas.height;\n _points = [];\n _setStyle(context.createLinearGradient(0, 0, 0, 300));\n _setPoints(n);\n },\n _play = function(t, angle) {\n if(!_isPlay) {\n _timerID = setInterval(function() {\n _render(angle)\n }, t);\n _isPlay = true;\n }\n },\n _stop = function() {\n if(_isPlay) {\n clearInterval(_timerID);\n _isPlay = false;\n }\n },\n // private methods\n _setStyle = function(grad) {\n grad.addColorStop(0, 'rgb(0, 0, 0)');\n grad.addColorStop(0.7, 'rgb(16, 16, 16)');\n grad.addColorStop(0.8, 'rgb(32, 32, 32)');\n grad.addColorStop(0.9, 'rgb(48, 48, 48)');\n grad.addColorStop(1, 'rgb(64, 64, 64)');\n _context.fillStyle = grad;\n _context.strokeStyle = 'rgb(0, 128, 255)';\n _context.fillRect(0, 0, _width, _height);\n },\n _setPoints = function(num) {\n var x, y, u, v, r, n = 0, point;\n for(var i=0; i<=2*Math.PI*num; i++) {\n for(var j=0; j<=2*Math.PI*4.8; j++) {\n u = i/num; v = j/4.8;\n r = 4*(1 - Math.cos(u)/2);\n if(u> 1),\n y = (point3d[1]*scale) + (_height >> 1);\n _context.lineWidth= scale << 1;\n _context.beginPath();\n _context.moveTo(x, y);\n _context.lineTo(x + scale, y);\n _context.stroke();\n },\n _rotateX = function(point3d, angle) {\n var y = point3d[1],\n z = point3d[2],\n c = Math.cos(angle),\n s = Math.sin(angle),\n ty = y,\n tz = z;\n y = ty*c - tz*s;\n z = ty*s + tz*c;\n point3d[1] = y;\n point3d[2] = z;\n },\n _rotateY = function(point3d, angle) {\n var x = point3d[0],\n z = point3d[2],\n c = Math.cos(angle),\n s = Math.sin(angle),\n tz = z,\n tx = x;\n x = tx*c + tz*s;\n z = tz*c - tx*s;\n point3d[0] = x;\n point3d[2] = z;\n },\n _rotateZ = function(point3d, angle) {\n var x = point3d[0],\n y = point3d[1],\n c = Math.cos(angle),\n s = Math.sin(angle),\n tx = x,\n ty = y;\n x = tx*c - ty*s;\n y = tx*s + ty*c;\n point3d[0] = x;\n point3d[1] = y;\n };\n // public APIs\n return {\n init: _init,\n play: _play,\n stop: _stop,\n };\n})();" }, { "path": "draw/lib3d.js", "content": "/**\n * My 3D Library v.0.0.2\n * \n */\nvar Class = {\n create : function() {\n var properties = arguments[0];\n function self() {\n this.initialize.apply(this, arguments);\n }\n for(var i in properties) {\n self.prototype[i] = properties[i];\n }\n if(!self.prototype.initialize) {\n self.prototype.initialize = function() {};\n }\n return self;\n }\n};\n\nvar Point = Class.create({\n initialize : function(x, y) {\n this.x = x || 0.0;\n this.y = y || 0.0;\n }\n});\n\nvar Vertex3D = Class.create({\n initialize : function(x, y, z) {\n this.x = x || 0.0;\n this.y = y || 0.0;\n this.z = z || 0.0;\n this.length = Math.sqrt(this.x*this.x + this.y*this.y + this.z*this.z);\n },\n clone : function() {\n return new Vertex3D(this.x, this.y, this.z);\n },\n add : function(v) {\n return new Vertex3D(this.x + v.x, this.y + v.y, this.z + v.z);\n },\n sub : function(v) {\n return new Vertex3D(this.x - v.x, this.y - v.y, this.z - v.z);\n },\n dot : function(v) {\n return this.x*v.x + this.y*v.y + this.z*v.z;\n },\n cross : function(v) {\n return new Vertex3D(v.y*this.z - v.z * this.y,\n v.z*this.x - v.x*this.z,\n v.x*this.y - v.y*this.x);\n },\n negate : function() {\n this.x *= -1;\n this.y *= -1;\n this.z *= -1;\n },\n normalize : function() {\n if(this.length != 0 && this.length != 1) {\n var mod = 1/this.length;\n this.x *= mod;\n this.y *= mod;\n this.z *= mod;\n }\n },\n project : function(screen) {\n var scale = screen.focalLength/(screen.focalLength + this.z);\n return new Point(\n this.x*scale + screen.projectionCenter.x,\n -this.y*scale + screen.projectionCenter.y\n );\n },\n scaleBy : function(s) {\n this.x *= s;\n this.y *= s;\n this.z *= s;\n },\n rotateX : function(angle) {\n var s = Math.sin(angle);\n var c = Math.cos(angle);\n return new Vertex3D(\n this.x,\n c*this.y - s*this.z,\n s*this.y + c*this.z\n );\n },\n rotateY : function(angle) {\n var s = Math.sin(angle);\n var c = Math.cos(angle);\n return new Vertex3D(\n s*this.z + c*this.x,\n this.y,\n c*this.z - s*this.x\n );\n },\n rotateZ : function(angle) {\n var s = Math.sin(angle);\n var c = Math.cos(angle);\n return new Vertex3D(\n c*this.x - s*this.y,\n s*this.x + c*this.y,\n this.z\n );\n }\n});\n\nvar TriangleMesh3D = Class.create({\n initialize : function(material, vertices, faces) {\n this.material = material;\n if(vertices && vertices.length == 3) {\n this.v0 = vertices[0];\n this.v1 = vertices[1];\n this.v2 = vertices[2];\n }else {\n this.v0 = new Vertex3D(1, 0, 0);\n this.v1 = new Vertex3D(0, 1, 0);\n this.v2 = new Vertex3D(0, 0, 1); \n }\n this.faces = faces || [];\n },\n draw : function(screen) {\n this.material.drawTriangle(this, screen);\n },\n rotateX : function(angle) {\n this.v0 = this.v0.rotateX(angle);\n this.v1 = this.v1.rotateX(angle);\n this.v2 = this.v2.rotateX(angle);\n },\n rotateY : function(angle) {\n this.v0 = this.v0.rotateY(angle);\n this.v1 = this.v1.rotateY(angle);\n this.v2 = this.v2.rotateY(angle);\n },\n rotateZ : function(angle) {\n this.v0 = this.v0.rotateZ(angle);\n this.v1 = this.v1.rotateZ(angle);\n this.v2 = this.v2.rotateZ(angle);\n },\n translate : function(v) {\n this.v0 = this.v0.add(v);\n this.v1 = this.v1.add(v);\n this.v2 = this.v2.add(v);\n }\n});\n\nvar Quaternion = Class.create({\n initialize : function(x, y, z, w) {\n this.x = x || 0.0;\n this.y = y || 0.0;\n this.z = z || 0.0;\n this.w = w || 1.0;\n this.EPSILON = 0.000001;\n this.DEGTORAD = Math.PI/180.0;\n this.RADTODEG = 180.0/Math.PI;\n },\n clone : function() {\n return new Quaternion(this.x, this.y, this.z, this.w);\n },\n calculateMultiply : function(a, b) {\n this.x = a.w*b.x + a.x*b.w + a.y*b.z - a.z*b.y;\n this.y = a.w*b.y - a.x*b.z + a.y*b.w + a.z*b.x;\n this.z = a.w*b.z + a.x*b.y - a.y*b.x + a.z*b.w;\n this.w = a.w*b.w - a.x*b.x - a.y*b.y - a.z*b.z;\n },\n setFromAxisAngle : function(x, y, z, angle) {\n var s = Math.sin(angle*0.5);\n var c = Math.cos(angle*0.5);\n this.x = x*s;\n this.y = y*s;\n this.z = z*s;\n this.w = c;\n this.normalize();\n },\n setFromEuler : function(ax, ay, az, useDegrees) {\n if(useDegrees) {\n ax *= DEGTORAD;\n ay *= DEGTORAD;\n az *= DEGTORAD;\n }\n var fSinPitch = Math.sin(ax*0.5);\n var fCosPitch = Math.cos(ax*0.5);\n var fSinYaw = Math.sin(ay*0.5);\n var fCosYaw = Math.cos(ay*0.5);\n var fSinRoll = Math.sin(az*0.5);\n var fCosRoll = Math.cos(az*0.5);\n var fCosPitchCosYaw = fCosPitch*fCosYaw;\n var fSinPitchSinYaw = fSinPitch*fSinYaw;\n this.x = fSinRoll*fCosPitchCosYaw - fCosRoll*fSinPitchSinYaw;\n this.y = fCosRoll*fSinPitch*fCosYaw + fSinRoll*fCosPitch*fSinYaw;\n this.z = fCosRoll*fCosPitch*fSinYaw - fSinRoll*fSinPitch*fCosYaw;\n this.w = fCosRoll*fCosPitchCosYaw + fSinRoll*fSinPitchSinYaw;\n },\n modulo : function() {\n return Math.sqrt(this.x*this.x + this.y*this.y + this.z*this.z * this.w*this.w);\n },\n conjugate : function(a) {\n return new Quaternion(-a.x, -a.y, -a.z, a.w);\n },\n createFromAxisAngle : function(ax, ay, az, useDegrees) {\n if(useDegrees) {\n ax *= DEGTORAD;\n ay *= DEGTORAD;\n az *= DEGTORAD;\n }\n var fSinPitch = Math.sin(ax*0.5);\n var fCosPitch = Math.cos(ax*0.5);\n var fSinYaw = Math.sin(ay*0.5);\n var fCosYaw = Math.cos(ay*0.5);\n var fSinRoll = Math.sin(az*0.5);\n var fCosRoll = Math.cos(az*0.5);\n var fCosPitchCosYaw = fCosPitch*fCosYaw;\n var fSinPitchSinYaw = fSinPitch*fSinYaw;\n var q = new Quaternion();\n q.x = fSinRoll*fCosPitchCosYaw - fCosRoll*fSinPitchSinYaw;\n q.y = fCosRoll*fSinPitch*fCosYaw + fSinRoll*fCosPitch*fSinYaw;\n q.z = fCosRoll*fCosPitch*fSinYaw - fSinRoll*fSinPitch*fCosYaw;\n q.w = fCosRoll*fCosPitchCosYaw + fSinRoll*fSinPitchSinYaw;\n return q;\n },\n add : function(a, b) {\n return new Quaternion(a.x + b.x, a.y + b.y, a.z + b.z, a.w + b.w);\n },\n sub : function(a, b) {\n return new Quaternion(a.x - b.x, a.y - b.y, a.z - b.z, a.w - b.w);\n },\n dot : function(a, b) {\n return (a.x*b.x) + (a.y*b.y) + (a.z*b.z) + (a.w*b.w);\n },\n multiply : function(a, b) {\n var c = new Quaternion();\n c.x = a.w*b.x + a.x*b.w + a.y*b.z - a.z*b.y;\n c.y = a.w*b.y - a.x*b.z + a.y*b.w + a.z*b.x;\n c.z = a.w*b.z + a.x*b.y - a.y*b.x + a.z*b.w;\n c.w = a.w*b.w - a.x*b.x - a.y*b.y - a.z*b.z;\n return c;\n },\n mult : function(b) {\n var ax = this.x;\n var ay = this.y;\n var az = this.z;\n var aw = this.w;\n this.x = aw*b.x + ax*b.w + ay*b.z - az*b.y;\n this.y = aw*b.y - ax*b.z + ay*b.w + az*b.x;\n this.z = aw*b.z + ax*b.y - ay*b.x + az*b.w;\n this.w = aw*b.w - ax*b.x - ay*b.y - az*b.z;\n },\n normalize : function() {\n var len = this.modulo();\n if(Math.abs(len) < this.EPSILON) {\n this.x = this.y = this.z = 0.0;\n this.w = 1.0;\n }else {\n var m = 1/len;\n this.x *= m;\n this.y *= m;\n this.z *= m;\n this.w *= m;\n }\n },\n slerp : function(qa, qb, alpha) {\n var angle = qa.w*qb.w + qa.x*qb.x + qa.y*qb.y + qa.z*qb.z;\n if(angle < 0.0) {\n qa.x *= -1.0;\n qa.y *= -1.0;\n qa.z *= -1.0;\n qa.w *= -1.0;\n angle *= -1.0;\n }\n var scale;\n var invscale;\n if((angle + 1.0) > this.EPSILON) {\n if((1.0 - angle) >= this.EPSILON) {\n var theta = Math.acos(angle);\n var invsintheta = 1.0/Math.sin(theta);\n scale = Math.sin(theta*(1.0 - alpha))*invsintheta;\n invscale = Math.sin(theta*alpha)*invsintheta;\n }else {\n scale = 1.0 - alpha;\n invscale = alpha;\n }\n }else {\n qb.y = -qa.y;\n qb.x = qa.x;\n qb.w = -qa.w;\n qb.z = qa.z;\n scale = Math.sin(Math.PI*(0.5 - alpha));\n invscale = Math.sin(Math.PI*alpha);\n }\n return new Quaternion(scale*qa.x + invscale*qb.x, \n scale*qa.y + invscale*qb.y,\n scale*qa.z + invscale*qb.z,\n scale*qa.w + invscale*qb.w );\n },\n toEuler : function() {\n var euler = {x:0, y:0, z:0};\n var q = this;\n var test = q.x*q.y + q.z*q.w;\n if(test > 0.499) {\n euler.x = 2*Math.atan2(q.x, q.w);\n euler.y = Math.PI/2;\n euler.z = 0;\n return euler;\n }\n if(test < -0.499) {\n euler.x = -2*Math.atan2(q.x, q.w);\n euler.y = -Math.PI/2;\n euler.z = 0;\n return euler;\n }\n var sqx = q.x*q.x;\n var sqy = q.y*q*y;\n var sqz = q.z*q.z;\n euler.x = Math.atan2(2*q.y*q.w - 2*q.x*q.z, 1 - 2*sqy - 2*sqz);\n euler.y = Math.asin(2*test);\n euler.z = Math.atan2(2*q.x*q.w - 2*q.y*q.z , 1 - 2*sqx - 2*sqz);\n return euler;\n }\n});\n\nvar Scene3D = Class.create({\n initialize : function() {\n this.objects = [];\n this.materials = [];\n },\n addObject : function(object) {\n this.objects.push(object);\n },\n removeObject : function(object) {\n for(var i=0; i> 1),\n y = (point3d[1]*scale) + (_height >> 1);\n _context.lineWidth= scale << 1;\n _context.beginPath();\n _context.moveTo(x, y);\n _context.lineTo(x + scale, y);\n _context.stroke();\n },\n _rotateX = function(point3d, angle) {\n var y = point3d[1],\n z = point3d[2],\n c = Math.cos(angle),\n s = Math.sin(angle),\n ty = y,\n tz = z;\n y = ty*c - tz*s;\n z = ty*s + tz*c;\n point3d[1] = y;\n point3d[2] = z;\n },\n _rotateY = function(point3d, angle) {\n var x = point3d[0],\n z = point3d[2],\n c = Math.cos(angle),\n s = Math.sin(angle),\n tz = z,\n tx = x;\n x = tx*c + tz*s;\n z = tz*c - tx*s;\n point3d[0] = x;\n point3d[2] = z;\n },\n _rotateZ = function(point3d, angle) {\n var x = point3d[0],\n y = point3d[1],\n c = Math.cos(angle),\n s = Math.sin(angle),\n tx = x,\n ty = y;\n x = tx*c - ty*s;\n y = tx*s + ty*c;\n point3d[0] = x;\n point3d[1] = y;\n };\n\n // public APIs\n return {\n init: _init,\n play: _play,\n stop: _stop,\n };\n})();" }, { "path": "draw/webgl/glsample.js", "content": "/**\n * WebGL Tutorial\n * https://developer.mozilla.org/ja/docs/Web/API/WebGL_API\n */\n\nvar gl; // WebGL コンテキスト用のグローバル変数\n\nfunction start() {\n var canvas = document.getElementById(\"Canvas\");\n\n // GL コンテキストを初期化\n gl = initWebGL(canvas);\n\n // WebGL を使用できる場合に限り、処理を継続\n\n if (gl) {\n // クリアカラーを黒色、不透明に設定する\n gl.clearColor(0.0, 0.0, 0.0, 1.0);\n // 深度テストを有効化\n gl.enable(gl.DEPTH_TEST);\n // 近くにある物体は、遠くにある物体を覆い隠す\n gl.depthFunc(gl.LEQUAL);\n // カラーバッファや深度バッファをクリアする\n gl.clear(gl.COLOR_BUFFER_BIT | gl.DEPTH_BUFFER_BIT);\n }\n}\n\nfunction initWebGL(canvas) {\n gl = null;\n\n try {\n // 標準コンテキストの取得を試みる。失敗した場合は、experimental にフォールバックする。\n gl = canvas.getContext(\"webgl\") || canvas.getContext(\"experimental-webgl\");\n } catch(e) {\n console.log(e);\n }\n\n // GL コンテキストを取得できない場合は終了する\n if (!gl) {\n alert(\"WebGL を初期化できません。ブラウザはサポートしていないようです。\");\n gl = null;\n }\n\n return gl;\n}" }, { "path": "draw/webgl/webglsample.html", "content": "\n\n \n \n \n WebGL Sample\n \n \n
\n

WebGL Sample

\n
\n
\n \n
\n \n \n \n\n" }, { "path": "ml/README.md", "content": "# Machine Learning Module\n\n## description\n\nMachine Learning in JavaScript / TypeScript (+ AngularJS)\n\n### list\n- [AROW](https://github.com/wellflat/imageprocessing-labs/tree/master/ml/arow)\n- [Decision Tree](https://github.com/wellflat/imageprocessing-labs/blob/master/ml/decisiontree/decisiontree.js)\n- [Denoising Autoencoders](https://github.com/wellflat/imageprocessing-labs/tree/master/ml/autoencoder)\n- [K-Means++](https://github.com/wellflat/imageprocessing-labs/blob/master/ml/kmeans/kmeans.js)\n- [Logistic Regression](https://github.com/wellflat/imageprocessing-labs/blob/master/ml/logistic_regression/logistic_regression.ts)\n- [SCW](https://github.com/wellflat/imageprocessing-labs/tree/master/ml/scw)\n- [Gradient Boosting(GBDT)](https://github.com/wellflat/imageprocessing-labs/tree/master/ml/gbdt)\n- [t-Distributed Stochastic Neighbor Embedding(t-SNE)](https://github.com/wellflat/imageprocessing-labs/tree/master/ml/t-sne)\n\n### sample\n[![da](https://github.com/wellflat/imageprocessing-labs/blob/master/images/da_demo.png)](https://rest-term.com/labs/html5/da/)\n\n[![pixel_clustering](https://github.com/wellflat/imageprocessing-labs/blob/master/images/pixel_clustering.jpg)](https://rest-term.com/labs/html5/pixelclustering.html)\n\n[![decision_tree](https://github.com/wellflat/imageprocessing-labs/blob/master/images/decision_tree.png)](https://rest-term.com/labs/html5/dtree.html)\n\n[![t-sne](https://github.com/wellflat/imageprocessing-labs/blob/master/images/tsne.png)](https://rest-term.com/labs/html5/tsne/)\n\n(Tests: Firefox75.0+, Chrome81.0+)\n\n### web sample\n[Denoising Autoencoders][DenoisingAutoencoders]\n\n[Pixel Clustering using K-Means++][PixelClustering]\n\n[Decision Tree Learning][DecisionTree]\n\n[t-Distributed Stochastic Neighbor Embedding(t-SNE)][tSNE]\n\nlicense\n----------\nCopyright © 2014 wellflat Licensed under the [MIT License][MIT]\n\n[DenoisingAutoencoders]: http://rest-term.com/labs/html5/da/\n[PixelClustering]: http://rest-term.com/labs/html5/pixelclustering.html\n[DecisionTree]: http://rest-term.com/labs/html5/dtree.html\n[tSNE]: https://rest-term.com/labs/html5/tsne/\n[MIT]: http://www.opensource.org/licenses/mit-license.php\n" }, { "path": "ml/arow/README.md", "content": "# Adaptive Regularization of Weight Vectors (AROW)\n\n* AROW algorithm and LIBSVM format data loader modules \n(JavaScript codes in 'out' directory)\n* LIBSVM Data: Classification (Binary Class) [news20.binary](https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/binary.html#news20.binary)\n\n## usage in TypeScript\n\n```typescript\nimport {Feature, DataSet} from './types';\nimport {AROW} from './arow';\nimport {DataLoader} from './data_loader';\n\nconst featureSize: number = 1355191;\nconst r: number = 0.1;\nconst clf: AROW = new AROW(featureSize, r);\nconst trainDataFile: string = 'data/news20_train';\nconst testDataFile: string = 'data/news20_test';\n\nconsole.log('asynchronous stream data read training');\nDataLoader.read(trainDataFile, (x: Feature, label: number) => {\n clf.update(x, label);\n}, () => {\n console.log('a iteration training complete.');\n let size: number = 0;\n let error: number = 0;\n\n DataLoader.read(testDataFile, (x: Feature, label: number) => {\n size++;\n let predLabel: number = clf.predict(x);\n if (predLabel != label) {\n error++;\n }\n }, () => {\n console.log('error rate = ' + (error / size));\n });\n});\n\nconsole.log('synchronous batch data read training');\nconst trainData: DataLoader = new DataLoader(trainDataFile);\nconst testData: DataLoader = new DataLoader(testDataFile);\nconsole.log('load data complete.');\nconsole.log(trainData.size, testData.size);\n\nconst maxIter: number = 5;\nfor (let i = 0; i < maxIter; i++) {\n trainData.data.forEach(e => {\n clf.update(e.x, e.label);\n });\n var error: number = 0;\n testData.data.forEach(e => {\n var predLabel: number = clf.predict(e.x);\n if (predLabel != e.label) {\n error++;\n }\n });\n console.log('iteration: ' + (i + 1) + ', error rate = ' + (error / testData.size));\n}\n```\n" }, { "path": "ml/arow/app.ts", "content": "import {Feature, DataSet} from './types';\nimport {AROW} from './arow';\nimport {DataLoader} from './data_loader';\n\nconst featureSize: number = 1355191;\nconst r: number = 0.1;\nconst clf: AROW = new AROW(featureSize, r);\nconst trainDataFile: string = 'data/news20_train';\nconst testDataFile: string = 'data/news20_test';\n\nconsole.log('asynchronous stream data read training');\nDataLoader.read(trainDataFile, (x: Feature, label: number) => {\n clf.update(x, label);\n}, () => {\n console.log('a iteration training complete.');\n let size: number = 0;\n let error: number = 0;\n\n DataLoader.read(testDataFile, (x: Feature, label: number) => {\n size++;\n let predLabel: number = clf.predict(x);\n if (predLabel != label) {\n error++;\n }\n }, () => {\n console.log('error rate = ' + (error / size));\n });\n});\n\nconsole.log('synchronous batch data read training');\nconst trainData: DataLoader = new DataLoader(trainDataFile);\nconst testData: DataLoader = new DataLoader(testDataFile);\nconsole.log('load data complete.');\nconsole.log(trainData.size, testData.size);\n\nconst maxIter: number = 5;\nfor (let i = 0; i < maxIter; i++) {\n trainData.data.forEach(e => {\n clf.update(e.x, e.label);\n });\n var error: number = 0;\n testData.data.forEach(e => {\n var predLabel: number = clf.predict(e.x);\n if (predLabel != e.label) {\n error++;\n }\n });\n console.log('iteration: ' + (i + 1) + ', error rate = ' + (error / testData.size));\n}\n" }, { "path": "ml/arow/arow.ts", "content": "/**\n * AROW (Adaptive Regularization of Weight Vectors) class\n */\n\nimport {Feature, DataSet} from './types';\n\nexport class AROW {\n private mean: Float32Array;\n private cov: Float32Array;\n private f: number;\n private r: number;\n\n constructor(featureSize: number, r: number = 0.1) {\n this.f = featureSize;\n this.mean = new Float32Array(featureSize);\n this.cov = new Float32Array(featureSize);\n this.cov.fill(1.0);\n this.r = r;\n }\n\n public predict(x: Feature): number {\n return this.computeMargin(x) > 0 ? 1 : -1;\n }\n\n public clear(): void {\n this.mean = new Float32Array(this.f);\n this.cov = new Float32Array(this.f);\n this.cov.fill(1.0);\n }\n\n public update(x: Feature, label: number): number {\n let margin: number = this.computeMargin(x);\n let loss: number = margin * label < 0 ? 1 : 0;\n if (margin * label >= 1) {\n return 0;\n }\n let confidence: number = this.computeConfidence(x);\n let beta: number = 1.0 / (confidence + this.r);\n let alpha: number = (1.0 - label * margin) * beta;\n x.forEach(e => {\n this.mean[e.index] += alpha * label * this.cov[e.index] * e.value;\n })\n x.forEach(e => {\n this.cov[e.index] = 1.0 / ((1.0 / this.cov[e.index]) + e.value * e.value / this.r);\n });\n return loss;\n }\n\n private computeMargin(x: Feature): number {\n let res:number = 0.0;\n x.forEach(e => res += this.mean[e.index] * e.value);\n return res;\n }\n\n private computeConfidence(x: Feature): number {\n let res:number = 0.0;\n x.forEach(e => res += this.cov[e.index] * e.value * e.value);\n return res;\n }\n}" }, { "path": "ml/arow/data_loader.ts", "content": "/**\n * Data loader for LIBSVM data format\n *