javacv人脸识别+人脸匹配

1、先安装一个opencv3，https://opencv.org/releases.html，安装好后里面有很多自带的检测器，

2、上代码，eclipse/idea可以不用配置opencv，按pom导入相关包即可

pom.xml

  4.0.0
  com.cy
  JavaCV_Learn_001
  0.0.1-SNAPSHOT
  JavaCV_Learn_001
  JavaCV_Learn_001
  
  
	
	
	    org.bytedeco
	    javacv
	    1.4.4
	
	
	
	    org.bytedeco
	    javacv-platform
	    1.4.4
	                                   
  

代码

package com.cy;

import static org.bytedeco.javacpp.opencv_highgui.imshow;
import static org.bytedeco.javacpp.opencv_highgui.waitKey;
import static org.bytedeco.javacpp.opencv_imgcodecs.imread;
import static org.bytedeco.javacpp.opencv_imgcodecs.imwrite;

import java.io.File;
import java.util.ArrayList;
import java.util.List;

import javax.swing.WindowConstants;

import org.bytedeco.javacpp.opencv_core;
import org.bytedeco.javacpp.opencv_core.Mat;
import org.bytedeco.javacpp.opencv_imgproc;
import org.bytedeco.javacpp.opencv_objdetect.CascadeClassifier;
import org.bytedeco.javacv.CanvasFrame;
import org.bytedeco.javacv.Frame;
import org.bytedeco.javacv.OpenCVFrameConverter;
import org.bytedeco.javacv.VideoInputFrameGrabber;
import org.opencv.core.Core;
import org.opencv.core.MatOfFloat;
import org.opencv.core.MatOfInt;
import org.opencv.imgcodecs.Imgcodecs;
import org.opencv.imgproc.Imgproc;

public class TestLoad {
	static {
		System.load("D:\\soft\\openCV3\\opencv\\build\\java\\x64\\opencv_java345.dll");
	}
	/**
	 * 测试javacv
	 * 
Title: testLoadImg
  
	 * 
Description: 
	 * 
	 * @author 陈阳  
	 * @date 2019年1月25日
	 */
	public static void testLoadImg() {
        //读取原始图片
        Mat image = imread("C:\\Users\\Administrator\\Desktop\\\\me\\Fri Jan 25 15-30-22.bmp"); 
        if (image.empty()) {
            System.err.println("加载图片出错，请检查图片路径！");
            return;
        }
        //显示图片 
        imshow("显示原始图像", image);

        //无限等待按键按下
        waitKey(0);
    }
	
	/**
	 * 测试摄像头
	 * 
Title: testCamera
  
	 * 
Description: 
	 * @author 陈阳  
	 * @date 2019年1月25日
	 * @throws Exception
	 * @throws InterruptedException
	 */
	public static void testCamera() throws Exception, InterruptedException
	{
		VideoInputFrameGrabber grabber = VideoInputFrameGrabber.createDefault(1);
        grabber.start();
        CanvasFrame canvasFrame = new CanvasFrame("摄像头");
        canvasFrame.setDefaultCloseOperation(WindowConstants.EXIT_ON_CLOSE);
        canvasFrame.setAlwaysOnTop(true);
        while (true) {
            if (!canvasFrame.isDisplayable()) {
                grabber.stop();
                System.exit(-1);
            }
            Frame frame = grabber.grab();
            canvasFrame.showImage(frame);
            Thread.sleep(30);
        }
	}
	
	/**
	 * 把人像图片的人脸识别并裁剪出来，保存到同目录
	 * 
Title: testFace
  
	 * 
Description: 
	 * @author 陈阳  
	 * @date 2019年1月26日
	 * @param path
	 * @throws Exception
	 * @throws InterruptedException
	 */
	public static void testFace(String path) throws Exception, InterruptedException
	{
		Mat image = imread(path); 
        if (image.empty()) {
            System.err.println("加载图片出错，请检查图片路径！");
            return;
        }
//        imshow("显示原始图像", image);
        
        opencv_core.RectVector faces = new opencv_core.RectVector();
        
      //加载检测器
  		CascadeClassifier face_cascade = new CascadeClassifier("D:\\soft\\openCV3\\opencv\\sources\\data\\haarcascades_cuda\\haarcascade_frontalface_alt.xml");//初始化人脸检测器
        
      //当前帧图片进行灰度+直方均衡
        Mat videoMatGray = new Mat();
        opencv_imgproc.cvtColor(image, videoMatGray, Imgproc.COLOR_BGRA2GRAY);
        opencv_imgproc.equalizeHist(videoMatGray, videoMatGray);
        
        //使用检测器进行检测，把结果放进集合中
        face_cascade.detectMultiScale(videoMatGray, faces);
        
        //把所有人脸数据绘制到图片中
        File imagePath = new File(path);
        String dir = imagePath.getParent();
        String name = imagePath.getName().substring(0, imagePath.getName().lastIndexOf("."));
        for (int i = 0; i < faces.size(); i++) {
            opencv_core.Rect face = faces.get(i);
            Mat img_region = new Mat(image, face);
//            imshow("人脸裁剪"+i, img_region);
            imwrite(dir+File.separator+name+" face"+i+".png", img_region);
        }
//        waitKey(0);
	}
	
	/**
	 * 测试自带人脸识别，视频识别
	 * 
Title: detectFace
  
	 * 
Description: 
	 * @author 陈阳  
	 * @date 2019年1月25日
	 * @throws org.bytedeco.javacv.FrameGrabber.Exception
	 * @throws InterruptedException
	 */
	public static void detectFace() throws org.bytedeco.javacv.FrameGrabber.Exception, InterruptedException
	{
		//连接摄像头
		VideoInputFrameGrabber grabber = VideoInputFrameGrabber.createDefault(1);
        grabber.start();
        
		OpenCVFrameConverter.ToMat convertToMat = new OpenCVFrameConverter.ToMat();
		//加载检测器
		CascadeClassifier face_cascade = new CascadeClassifier("D:\\soft\\openCV3\\opencv\\sources\\data\\haarcascades_cuda\\haarcascade_frontalface_alt.xml");//初始化人脸检测器
		CascadeClassifier eye_cascade = new CascadeClassifier("D:\\soft\\openCV3\\opencv\\sources\\data\\haarcascades_cuda\\haarcascade_eye_tree_eyeglasses.xml");//初始化眼部检测器
        //定义人脸集合，矩形集合
		opencv_core.RectVector faces = new opencv_core.RectVector();
		opencv_core.RectVector eyes = new opencv_core.RectVector();
		
        CanvasFrame canvas = new CanvasFrame("人脸识别",1);//新建一个窗口
        canvas.setDefaultCloseOperation(WindowConstants.EXIT_ON_CLOSE);
        canvas.setAlwaysOnTop(true);
        
        //未关闭就一直向窗口刷新图片。就是视频
        while (true) {
        	//
            if (!canvas.isVisible()) {
                break;
            }
            Frame frame = grabber.grab();
            
            //获取当前帧图片
            Mat mat = convertToMat.convert(frame);
            if (mat.empty())
                continue;
            //当前帧图片进行灰度+直方均衡
            Mat videoMatGray = new Mat();
            opencv_imgproc.cvtColor(mat, videoMatGray, Imgproc.COLOR_BGRA2GRAY);
            opencv_imgproc.equalizeHist(videoMatGray, videoMatGray);
            
            //使用检测器进行检测，把结果放进集合中
            face_cascade.detectMultiScale(videoMatGray, faces);
            eye_cascade.detectMultiScale(videoMatGray, eyes);
            
            //把所有人脸数据绘制到图片中
            for (int i = 0; i < faces.size(); i++) {
                opencv_core.Rect face = faces.get(i);
                opencv_imgproc.rectangle(mat, face, opencv_core.Scalar.RED, 1, 8, 0);
            }
            for (int i = 0; i < eyes.size(); i++) {
                opencv_core.Rect eye = eyes.get(i);
                opencv_imgproc.rectangle(mat, eye, opencv_core.Scalar.GREEN, 1, 8, 0);
            }

            //把图片刷新到窗口
            canvas.showImage(convertToMat.convert(mat));
            
            Thread.sleep(30);//30毫秒刷新一次图像
        }
	}
	
	/**
	 * 直方图比较两个图片是否相似，可以用来比较两个人脸图片是否是同一个人
	 * 
Title: CmpPic
  
	 * 
Description: 
	 * @author 陈阳  
	 * @date 2019年1月26日
	 * @param src
	 * @param des
	 * @return
	 */
	public static double CmpPic(String src, String des) {
		System.out.println("\n==========直方图比较==========");
		//自定义阈值
		//相关性阈值，应大于多少，越接近1表示越像，最大为1
		double HISTCMP_CORREL_THRESHOLD = 0.7;
		//卡方阈值，应小于多少，越接近0表示越像
		double HISTCMP_CHISQR_THRESHOLD = 2;
		//交叉阈值，应大于多少，数值越大表示越像
		double HISTCMP_INTERSECT_THRESHOLD = 1.2;
		//巴氏距离阈值，应小于多少，越接近0表示越像
		double HISTCMP_BHATTACHARYYA_THRESHOLD = 0.3;
		
		try {
 
			long startTime = System.currentTimeMillis();
 
			org.opencv.core.Mat mat_src = Imgcodecs.imread(src);
			org.opencv.core.Mat mat_des = Imgcodecs.imread(des);;
 
			if (mat_src.empty() || mat_des.empty()) {
				throw new Exception("no file.");
			}
 
			org.opencv.core.Mat hsv_src = new org.opencv.core.Mat();
			org.opencv.core.Mat hsv_des = new org.opencv.core.Mat();
 
			// 转换成HSV
			Imgproc.cvtColor(mat_src, hsv_src, Imgproc.COLOR_BGR2HSV);
			Imgproc.cvtColor(mat_des, hsv_des, Imgproc.COLOR_BGR2HSV);
 
			List listImg1 = new ArrayList();
			List listImg2 = new ArrayList();
			listImg1.add(hsv_src);
			listImg2.add(hsv_des);
 
			MatOfFloat ranges = new MatOfFloat(0, 255);
			MatOfInt histSize = new MatOfInt(50);
			MatOfInt channels = new MatOfInt(0);
 
			org.opencv.core.Mat histImg1 = new org.opencv.core.Mat();
			org.opencv.core.Mat histImg2 = new org.opencv.core.Mat();
 
			//org.bytedeco.javacpp中的方法不太了解参数，所以直接上org.opencv中的方法，所以需要加载一下dll，System.load("D:\\soft\\openCV3\\opencv\\build\\java\\x64\\opencv_java345.dll");
			//opencv_imgproc.calcHist(images, nimages, channels, mask, hist, dims, histSize, ranges, uniform, accumulate);
			Imgproc.calcHist(listImg1, channels, new org.opencv.core.Mat(), histImg1, histSize, ranges);
			Imgproc.calcHist(listImg2, channels, new org.opencv.core.Mat(), histImg2, histSize, ranges);
 
			org.opencv.core.Core.normalize(histImg1, histImg1, 0d, 1d, Core.NORM_MINMAX, -1,
					new org.opencv.core.Mat());
			org.opencv.core.Core.normalize(histImg2, histImg2, 0d, 1d, Core.NORM_MINMAX, -1,
					new org.opencv.core.Mat());
 
			double result0, result1, result2, result3;
			result0 = Imgproc.compareHist(histImg1, histImg2, Imgproc.HISTCMP_CORREL);
			result1 = Imgproc.compareHist(histImg1, histImg2, Imgproc.HISTCMP_CHISQR);
			result2 = Imgproc.compareHist(histImg1, histImg2, Imgproc.HISTCMP_INTERSECT);
			result3 = Imgproc.compareHist(histImg1, histImg2, Imgproc.HISTCMP_BHATTACHARYYA);
 
			System.out.println("相关性（度量越高，匹配越准确 [基准："+HISTCMP_CORREL_THRESHOLD+"]）,当前值:" + result0);
			System.out.println("卡方（度量越低，匹配越准确 [基准："+HISTCMP_CHISQR_THRESHOLD+"]）,当前值:" + result1);
			System.out.println("交叉核（度量越高，匹配越准确 [基准："+HISTCMP_INTERSECT_THRESHOLD+"]）,当前值:" + result2);
			System.out.println("巴氏距离（度量越低，匹配越准确 [基准："+HISTCMP_BHATTACHARYYA_THRESHOLD+"]）,当前值:" + result3);
 
			//一共四种方式，有三个满足阈值就算匹配成功
			int count = 0;
			if (result0 > HISTCMP_CORREL_THRESHOLD)
				count++;
			if (result1 < HISTCMP_CHISQR_THRESHOLD)
				count++;
			if (result2 > HISTCMP_INTERSECT_THRESHOLD)
				count++;
			if (result3 < HISTCMP_BHATTACHARYYA_THRESHOLD)
				count++;
			int retVal = 0;
			if (count >= 3) {
				//这是相似的图像
				retVal = 1;
			}
 
			long estimatedTime = System.currentTimeMillis() - startTime;
 
			System.out.println("花费时间= " + estimatedTime + "ms");
 
			return retVal;
		} catch (Exception e) {
			System.out.println("例外:" + e);
		}
		return 0;
    }
	
	public static void main(String[] args) {
		try {
//			detectFace();
//			testCamera();
//			testFace("C:\\Users\\Administrator\\Desktop\\face\\Sat Jan 26 10-58-41.bmp");
			System.out.println(CmpPic("C:\\Users\\Administrator\\Desktop\\face\\Sat Jan 26 10-58-41 face0.png", "C:\\Users\\Administrator\\Desktop\\face\\Sat Jan 26 10-57-38 face0.png"));
		} catch (Exception e) {
			// TODO Auto-generated catch block
			e.printStackTrace();
		}
	}
}

效果：

opencv自带的有很多人脸检测模型，比如当前这个对正脸效果比较好，歪的和侧脸效果就不太好，可以自己多试试，有意思~