1.读取本地视频流,pom依赖
依赖于 org.bytedeco下的javacv/opencv/ffmpeg 包
<dependency> <groupId>org.bytedeco</groupId> <artifactId>javacv</artifactId> <version>1.4.3</version> </dependency> <dependency> <groupId>org.bytedeco.javacpp-presets</groupId> <artifactId>opencv</artifactId> <version>3.4.3-1.4.3</version> <classifier>linux-x86_64</classifier> </dependency> <dependency> <groupId>org.bytedeco.javacpp-presets</groupId> <artifactId>ffmpeg</artifactId> <version>4.0.2-1.4.3</version> <classifier>linux-x86_64</classifier> </dependency>
2.读取本地视频流并解帧为 opencv_core.Mat
File file = new File("/home/lab/javacv/t11.mp4"); FFmpegFrameGrabber grabber = new FFmpegFrameGrabber(file); grabber.start(); // 解帧为opencv_core.Mat List<opencv_core.Mat> mats = new ArrayList<>(); for (int i = 0; i < grabber.getLengthInFrames(); i++) { Frame frame = grabber.grabImage(); OpenCVFrameConverter.ToMat toMat = new OpenCVFrameConverter.ToMat(); opencv_core.Mat mat = toMat.convert(frame); if (mat != null) { mats.add(mat.clone()); } } grabber.stop();
3.获取32位dhash特征
dhash特征提取思路,图片Mat转为单通道的灰度图,并重置为5*5的Size,最后将其转储为长度为 25 的byte数组用以求取32位dhash特征
// 声明空的灰度图 Mat opencv_core.Mat grayImg = new opencv_core.Mat(mat.rows(), mat.cols(), opencv_imgcodecs.IMREAD_GRAYSCALE); // 转储为灰度图 opencv_imgproc.cvtColor(mat, grayImg, opencv_imgproc.COLOR_RGB2GRAY); // 修改Mat长宽size opencv_core.Mat resizedImg = new opencv_core.Mat(); opencv_core.Size size = new opencv_core.Size(5,5); opencv_imgproc.resize(grayImg,resizedImg,size); // 转为 5*5 byte 数组 byte[] bytePixels = new byte[5 * 5]; resizedImg.data().get(bytePixels); int[] pixels = new int[bytePixels.length]; for (int i=0; i<pixels.length; i++) { pixels[i] = bytePixels[i] & 0xff; } // 获取32位dhash特征 int feature = 0; for (int j=0; j<4; j++) { for (int i=0; i<4; i++) { int colBit = pixels[i*5+j] > pixels[(i+1)*5+j] ? 1 : 0; feature = (feature << 1) + colBit; int rowBit = pixels[i*5+j] > pixels[i*5+j+1] ? 1 : 0 ; feature = (feature << 1) + rowBit; } }
多线程部分,可参考该博: https://www.cnblogs.com/nyatom/p/10119306.html