opencv java小应用：比较两个图片的相似度

package com.company;

import org.opencv.core.*;
import org.opencv.imgcodecs.Imgcodecs;
import org.opencv.imgproc.Imgproc;
import org.opencv.objdetect.CascadeClassifier;

import java.util.Arrays;

public class FaceCompareMain {

    //初始化人脸探测器
    static CascadeClassifier faceDetector;

    static {
        System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
        faceDetector = new CascadeClassifier(
            "D:\\ib\\face-detact\\src\\com\\company\\haarcascade_frontalface_alt.xml");
    }

    // 1.  灰度化（减小图片大小）
    // 2. 人脸识别
    // 3. 人脸切割
    // 4. 规一化(人脸直方图)
    // 5. 直方图相似度匹配

    public static void main(String[] args) {
        String basePicPath = "D:\\ib\\face-detact\\src\\pics\\";
        double compareHist = compare_image(basePicPath + "11_1.png", basePicPath + "11_2.png");
        System.out.println(compareHist);
        if (compareHist > 0.72) {
            System.out.println("人脸匹配");
        } else {
            System.out.println("人脸不匹配");
        }
    }

    public static double compare_image(String img_1, String img_2) {
        Mat mat_1 = conv_Mat(img_1);
        Mat mat_2 = conv_Mat(img_2);

        Mat hist_1 = new Mat();
        Mat hist_2 = new Mat();

        //颜色范围
        MatOfFloat ranges = new MatOfFloat(0f, 256f);
        //直方图大小， 越大匹配越精确 (越慢)
        MatOfInt histSize = new MatOfInt(1000);

        Imgproc.calcHist(Arrays.asList(mat_1), new MatOfInt(0), new Mat(), hist_1, histSize, ranges);
        Imgproc.calcHist(Arrays.asList(mat_2), new MatOfInt(0), new Mat(), hist_2, histSize, ranges);

        // CORREL 相关系数
        double res = Imgproc.compareHist(hist_1, hist_2, Imgproc.CV_COMP_CORREL);
        return res;
    }

    // "D:\\ib\\face-detact\\src\\com\\company\\a1.jpg"
    private static Mat conv_Mat(String img_1) {
        Mat image0 = Imgcodecs.imread(img_1);

        Mat image = new Mat();
        //灰度转换
        Imgproc.cvtColor(image0, image, Imgproc.COLOR_BGR2GRAY);

        MatOfRect faceDetections = new MatOfRect();
        //探测人脸
        faceDetector.detectMultiScale(image, faceDetections);

        // rect中是人脸图片的范围
        for (Rect rect : faceDetections.toArray()) {
            //切割rect人脸
            Mat mat = new Mat(image, rect);
            return mat;
        }
        return null;
    }

}

代码
本文使用opencv 3.4.5版本，opencv大版本api变动不少

java项目设置，需要引入opencv native动态连接库

python代码

"""
pip install scipy numpy Pillow sewar
"""
from sewar.full_ref import uqi
from PIL import Image
import numpy as np

img1 = Image.open('C:\\Users\\shuayan\\Pictures\\axa.jpg')
img2 = Image.open('C:\\Users\\shuayan\\Pictures\\axa2.jpg')

max_width = max(img1.width, img2.width)
max_height = max(img1.height, img2.height)

img1 = img1.resize((max_width, max_height))
img2 = img2.resize((max_width, max_height))

# uqi要求两个图片大小一样， img2图片可以经过模糊，杂色，放大，裁剪，旋转等处理
print(uqi(np.asarray(img1), np.asarray(img2)))

参考 ： https://github.com/opencv/opencv/releases