List of usage examples for org.opencv.core Core kmeans
public static double kmeans(Mat data, int K, Mat bestLabels, TermCriteria criteria, int attempts, int flags, Mat centers)
From source file:classes.TextRecognitionPreparer.java
public static Scalar cluster(Scalar userColor, Mat cutout, int k) { Mat samples = cutout.reshape(1, cutout.cols() * cutout.rows()); Mat samples32f = new Mat(); samples.convertTo(samples32f, CvType.CV_32F, 1.0 / 255.0); Mat labels = new Mat(); TermCriteria criteria = new TermCriteria(TermCriteria.COUNT, 100, 1); Mat centers = new Mat(); Core.kmeans(samples32f, k, labels, criteria, 1, Core.KMEANS_PP_CENTERS, centers); Scalar fillingColor = getFillingColor(userColor, cutout, labels, centers); return fillingColor; }
From source file:com.astrocytes.core.operationsengine.OperationsImpl.java
License:Open Source License
private Mat applyKmeans(Mat source) { Mat dest = new Mat(); source.convertTo(source, CvType.CV_32F, 1.0 / 255.0); Mat centers = new Mat(); Mat labels = new Mat(); TermCriteria criteria = new TermCriteria(TermCriteria.COUNT, 20, 0.1); Core.kmeans(source, 4, labels, criteria, 10, Core.KMEANS_PP_CENTERS, centers); List<Mat> mats = showClusters(source, labels, centers); //mats.get(0).convertTo(dest, CvType.CV_8UC3); Core.merge(mats, dest);//from w ww . j a v a 2 s . c o m //centers.convertTo(dest, CvType.CV_8UC3); return dest; }
From source file:nz.ac.auckland.lablet.vision.CamShiftTracker.java
License:Open Source License
/** * Finds the dominant colour in an image, and returns two values in HSV colour space to represent similar colours, * e.g. so you can keep all colours similar to the dominant colour. * * How the algorithm works://from w ww . j av a2 s .c o m * * 1. Scale the frame down so that algorithm doesn't take too long. * 2. Segment the frame into different colours (number of colours determined by k) * 3. Find dominant cluster (largest area) and get its central colour point. * 4. Get range (min max) to represent similar colours. * * @param bgr The input frame, in BGR colour space. * @param k The number of segments to use (2 works well). * @return The min and max HSV colour values, which represent the colours similar to the dominant colour. */ private Pair<Scalar, Scalar> getMinMaxHsv(Mat bgr, int k) { //Convert to HSV Mat input = new Mat(); Imgproc.cvtColor(bgr, input, Imgproc.COLOR_BGR2BGRA, 3); //Scale image Size bgrSize = bgr.size(); Size newSize = new Size(); if (bgrSize.width > CamShiftTracker.KMEANS_IMG_SIZE || bgrSize.height > CamShiftTracker.KMEANS_IMG_SIZE) { if (bgrSize.width > bgrSize.height) { newSize.width = CamShiftTracker.KMEANS_IMG_SIZE; newSize.height = CamShiftTracker.KMEANS_IMG_SIZE / bgrSize.width * bgrSize.height; } else { newSize.width = CamShiftTracker.KMEANS_IMG_SIZE / bgrSize.height * bgrSize.width; newSize.height = CamShiftTracker.KMEANS_IMG_SIZE; } Imgproc.resize(input, input, newSize); } //Image quantization using k-means, see here for details of k-means algorithm: http://bit.ly/1JIvrlB Mat clusterData = new Mat(); Mat reshaped = input.reshape(1, input.rows() * input.cols()); reshaped.convertTo(clusterData, CvType.CV_32F, 1.0 / 255.0); Mat labels = new Mat(); Mat centres = new Mat(); TermCriteria criteria = new TermCriteria(TermCriteria.COUNT, 50, 1); Core.kmeans(clusterData, k, labels, criteria, 1, Core.KMEANS_PP_CENTERS, centres); //Get num hits for each category int[] counts = new int[k]; for (int i = 0; i < labels.rows(); i++) { int label = (int) labels.get(i, 0)[0]; counts[label] += 1; } //Get cluster index with maximum number of members int maxCluster = 0; int index = -1; for (int i = 0; i < counts.length; i++) { int value = counts[i]; if (value > maxCluster) { maxCluster = value; index = i; } } //Get cluster centre point hsv int r = (int) (centres.get(index, 2)[0] * 255.0); int g = (int) (centres.get(index, 1)[0] * 255.0); int b = (int) (centres.get(index, 0)[0] * 255.0); int sum = (r + g + b) / 3; //Get colour range Scalar min; Scalar max; int rg = Math.abs(r - g); int gb = Math.abs(g - b); int rb = Math.abs(r - b); int maxDiff = Math.max(Math.max(rg, gb), rb); if (maxDiff < 35 && sum > 120) { //white min = new Scalar(0, 0, 0); max = new Scalar(180, 40, 255); } else if (sum < 50 && maxDiff < 35) { //black min = new Scalar(0, 0, 0); max = new Scalar(180, 255, 40); } else { Mat bgrColour = new Mat(1, 1, CvType.CV_8UC3, new Scalar(r, g, b)); Mat hsvColour = new Mat(); Imgproc.cvtColor(bgrColour, hsvColour, Imgproc.COLOR_BGR2HSV, 3); double[] hsv = hsvColour.get(0, 0); int addition = 0; int minHue = (int) hsv[0] - colourRange; if (minHue < 0) { addition = Math.abs(minHue); } int maxHue = (int) hsv[0] + colourRange; min = new Scalar(Math.max(minHue, 0), 60, Math.max(35, hsv[2] - 30)); max = new Scalar(Math.min(maxHue + addition, 180), 255, 255); } return new Pair<>(min, max); }
From source file:opencv.CaptchaDetection.java
private static Mat k_means_spilter(Mat src) { Mat dst = Mat.zeros(src.size(), CvType.CV_8UC1); int width = src.cols(); int height = src.rows(); int dims = src.channels(); // //from ww w. ja v a 2 s . c o m int clusterCount = 3; Mat points = new Mat(width * height, dims, CvType.CV_32F, new Scalar(0)); Mat centers = new Mat(clusterCount, dims, CvType.CV_32F); Mat labels = new Mat(width * height, 1, CvType.CV_32S); // points for (int row = 0; row < height; row++) { for (int col = 0; col < width; col++) { int index = row * width + col; double[] s_data = src.get(row, col); for (int channel = 0; channel < 3; channel++) { float[] f_buff = new float[1]; f_buff[0] = (float) s_data[channel]; points.put(index, channel, f_buff); } } } // knn ? TermCriteria criteria = new TermCriteria(TermCriteria.EPS + TermCriteria.MAX_ITER, 10, 0.1); Core.kmeans(points, clusterCount, labels, criteria, 3, Core.KMEANS_PP_CENTERS, centers); // ??? label index Map<Integer, Integer> tmp = new TreeMap<>(); for (int i = 0; i < clusterCount; i++) { int sum = 0; for (int j = 0; j < dims; j++) { sum += centers.get(i, j)[0]; } while (tmp.containsKey(sum)) sum++; tmp.put(sum, i); } int count = 0; int[] label_order = new int[clusterCount]; for (Map.Entry<Integer, Integer> iter : tmp.entrySet()) { label_order[count++] = iter.getValue(); } for (int row = 0; row < height; row++) { for (int col = 0; col < width; col++) { int index = row * width + col; int label = (int) labels.get(index, 0)[0]; if (label == label_order[1]) { byte[] d_buff = new byte[1]; d_buff[0] = (byte) 255; dst.put(row, col, d_buff); } } } return dst; }