1 import java.util.ArrayList; 2 import java.util.List; 3 import java.util.Random; 4 5 import org.opencv.core.Core; 6 import org.opencv.core.CvType; 7 import org.opencv.core.Mat; 8 import org.opencv.core.MatOfPoint; 9 import org.opencv.core.Point; 10 import org.opencv.core.Scalar; 11 import org.opencv.highgui.HighGui; 12 import org.opencv.imgcodecs.Imgcodecs; 13 import org.opencv.imgproc.Imgproc; 14 15 /** 16 * 17 * @brief Sample code showing how to segment overlapping objects using Laplacian filtering, in addition to Watershed 18 * and Distance Transformation 19 * 20 */ 21 class ImageSegmentation { run(String[] args)22 public void run(String[] args) { 23 //! [load_image] 24 // Load the image 25 String filename = args.length > 0 ? args[0] : "../data/cards.png"; 26 Mat srcOriginal = Imgcodecs.imread(filename); 27 if (srcOriginal.empty()) { 28 System.err.println("Cannot read image: " + filename); 29 System.exit(0); 30 } 31 32 // Show source image 33 HighGui.imshow("Source Image", srcOriginal); 34 //! [load_image] 35 36 //! [black_bg] 37 // Change the background from white to black, since that will help later to 38 // extract 39 // better results during the use of Distance Transform 40 Mat src = srcOriginal.clone(); 41 byte[] srcData = new byte[(int) (src.total() * src.channels())]; 42 src.get(0, 0, srcData); 43 for (int i = 0; i < src.rows(); i++) { 44 for (int j = 0; j < src.cols(); j++) { 45 if (srcData[(i * src.cols() + j) * 3] == (byte) 255 && srcData[(i * src.cols() + j) * 3 + 1] == (byte) 255 46 && srcData[(i * src.cols() + j) * 3 + 2] == (byte) 255) { 47 srcData[(i * src.cols() + j) * 3] = 0; 48 srcData[(i * src.cols() + j) * 3 + 1] = 0; 49 srcData[(i * src.cols() + j) * 3 + 2] = 0; 50 } 51 } 52 } 53 src.put(0, 0, srcData); 54 55 // Show output image 56 HighGui.imshow("Black Background Image", src); 57 //! [black_bg] 58 59 //! [sharp] 60 // Create a kernel that we will use to sharpen our image 61 Mat kernel = new Mat(3, 3, CvType.CV_32F); 62 // an approximation of second derivative, a quite strong kernel 63 float[] kernelData = new float[(int) (kernel.total() * kernel.channels())]; 64 kernelData[0] = 1; kernelData[1] = 1; kernelData[2] = 1; 65 kernelData[3] = 1; kernelData[4] = -8; kernelData[5] = 1; 66 kernelData[6] = 1; kernelData[7] = 1; kernelData[8] = 1; 67 kernel.put(0, 0, kernelData); 68 69 // do the laplacian filtering as it is 70 // well, we need to convert everything in something more deeper then CV_8U 71 // because the kernel has some negative values, 72 // and we can expect in general to have a Laplacian image with negative values 73 // BUT a 8bits unsigned int (the one we are working with) can contain values 74 // from 0 to 255 75 // so the possible negative number will be truncated 76 Mat imgLaplacian = new Mat(); 77 Imgproc.filter2D(src, imgLaplacian, CvType.CV_32F, kernel); 78 Mat sharp = new Mat(); 79 src.convertTo(sharp, CvType.CV_32F); 80 Mat imgResult = new Mat(); 81 Core.subtract(sharp, imgLaplacian, imgResult); 82 83 // convert back to 8bits gray scale 84 imgResult.convertTo(imgResult, CvType.CV_8UC3); 85 imgLaplacian.convertTo(imgLaplacian, CvType.CV_8UC3); 86 87 // imshow( "Laplace Filtered Image", imgLaplacian ); 88 HighGui.imshow("New Sharped Image", imgResult); 89 //! [sharp] 90 91 //! [bin] 92 // Create binary image from source image 93 Mat bw = new Mat(); 94 Imgproc.cvtColor(imgResult, bw, Imgproc.COLOR_BGR2GRAY); 95 Imgproc.threshold(bw, bw, 40, 255, Imgproc.THRESH_BINARY | Imgproc.THRESH_OTSU); 96 HighGui.imshow("Binary Image", bw); 97 //! [bin] 98 99 //! [dist] 100 // Perform the distance transform algorithm 101 Mat dist = new Mat(); 102 Imgproc.distanceTransform(bw, dist, Imgproc.DIST_L2, 3); 103 104 // Normalize the distance image for range = {0.0, 1.0} 105 // so we can visualize and threshold it 106 Core.normalize(dist, dist, 0.0, 1.0, Core.NORM_MINMAX); 107 Mat distDisplayScaled = new Mat(); 108 Core.multiply(dist, new Scalar(255), distDisplayScaled); 109 Mat distDisplay = new Mat(); 110 distDisplayScaled.convertTo(distDisplay, CvType.CV_8U); 111 HighGui.imshow("Distance Transform Image", distDisplay); 112 //! [dist] 113 114 //! [peaks] 115 // Threshold to obtain the peaks 116 // This will be the markers for the foreground objects 117 Imgproc.threshold(dist, dist, 0.4, 1.0, Imgproc.THRESH_BINARY); 118 119 // Dilate a bit the dist image 120 Mat kernel1 = Mat.ones(3, 3, CvType.CV_8U); 121 Imgproc.dilate(dist, dist, kernel1); 122 Mat distDisplay2 = new Mat(); 123 dist.convertTo(distDisplay2, CvType.CV_8U); 124 Core.multiply(distDisplay2, new Scalar(255), distDisplay2); 125 HighGui.imshow("Peaks", distDisplay2); 126 //! [peaks] 127 128 //! [seeds] 129 // Create the CV_8U version of the distance image 130 // It is needed for findContours() 131 Mat dist_8u = new Mat(); 132 dist.convertTo(dist_8u, CvType.CV_8U); 133 134 // Find total markers 135 List<MatOfPoint> contours = new ArrayList<>(); 136 Mat hierarchy = new Mat(); 137 Imgproc.findContours(dist_8u, contours, hierarchy, Imgproc.RETR_EXTERNAL, Imgproc.CHAIN_APPROX_SIMPLE); 138 139 // Create the marker image for the watershed algorithm 140 Mat markers = Mat.zeros(dist.size(), CvType.CV_32S); 141 142 // Draw the foreground markers 143 for (int i = 0; i < contours.size(); i++) { 144 Imgproc.drawContours(markers, contours, i, new Scalar(i + 1), -1); 145 } 146 147 // Draw the background marker 148 Mat markersScaled = new Mat(); 149 markers.convertTo(markersScaled, CvType.CV_32F); 150 Core.normalize(markersScaled, markersScaled, 0.0, 255.0, Core.NORM_MINMAX); 151 Imgproc.circle(markersScaled, new Point(5, 5), 3, new Scalar(255, 255, 255), -1); 152 Mat markersDisplay = new Mat(); 153 markersScaled.convertTo(markersDisplay, CvType.CV_8U); 154 HighGui.imshow("Markers", markersDisplay); 155 Imgproc.circle(markers, new Point(5, 5), 3, new Scalar(255, 255, 255), -1); 156 //! [seeds] 157 158 //! [watershed] 159 // Perform the watershed algorithm 160 Imgproc.watershed(imgResult, markers); 161 162 Mat mark = Mat.zeros(markers.size(), CvType.CV_8U); 163 markers.convertTo(mark, CvType.CV_8UC1); 164 Core.bitwise_not(mark, mark); 165 // imshow("Markers_v2", mark); // uncomment this if you want to see how the mark 166 // image looks like at that point 167 168 // Generate random colors 169 Random rng = new Random(12345); 170 List<Scalar> colors = new ArrayList<>(contours.size()); 171 for (int i = 0; i < contours.size(); i++) { 172 int b = rng.nextInt(256); 173 int g = rng.nextInt(256); 174 int r = rng.nextInt(256); 175 176 colors.add(new Scalar(b, g, r)); 177 } 178 179 // Create the result image 180 Mat dst = Mat.zeros(markers.size(), CvType.CV_8UC3); 181 byte[] dstData = new byte[(int) (dst.total() * dst.channels())]; 182 dst.get(0, 0, dstData); 183 184 // Fill labeled objects with random colors 185 int[] markersData = new int[(int) (markers.total() * markers.channels())]; 186 markers.get(0, 0, markersData); 187 for (int i = 0; i < markers.rows(); i++) { 188 for (int j = 0; j < markers.cols(); j++) { 189 int index = markersData[i * markers.cols() + j]; 190 if (index > 0 && index <= contours.size()) { 191 dstData[(i * dst.cols() + j) * 3 + 0] = (byte) colors.get(index - 1).val[0]; 192 dstData[(i * dst.cols() + j) * 3 + 1] = (byte) colors.get(index - 1).val[1]; 193 dstData[(i * dst.cols() + j) * 3 + 2] = (byte) colors.get(index - 1).val[2]; 194 } else { 195 dstData[(i * dst.cols() + j) * 3 + 0] = 0; 196 dstData[(i * dst.cols() + j) * 3 + 1] = 0; 197 dstData[(i * dst.cols() + j) * 3 + 2] = 0; 198 } 199 } 200 } 201 dst.put(0, 0, dstData); 202 203 // Visualize the final image 204 HighGui.imshow("Final Result", dst); 205 //! [watershed] 206 207 HighGui.waitKey(); 208 System.exit(0); 209 } 210 } 211 212 public class ImageSegmentationDemo { main(String[] args)213 public static void main(String[] args) { 214 // Load the native OpenCV library 215 System.loadLibrary(Core.NATIVE_LIBRARY_NAME); 216 217 new ImageSegmentation().run(args); 218 } 219 } 220
