OpenCV计算两幅图像的EMD度量

OpenCV计算两幅图像的EMD度量

EMD度量：

相关基础

https://blog.csdn.net/wangdonggg/article/details/32329879
这位大哥说的很好了，先就不搬运了，后面有空再整理

函数介绍

https://docs.opencv.org/4.5.2/d6/dc7/group__imgproc__hist.html

Python版本

看到现在有的都是C++版本的代码，自己基于Python写了一个

import cv2 as cv2
import numpy


def get_hist(img_in):
    img = img_in.copy()
    hsv = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
    # cv2.imshow("hsv", hsv)
    # calcHist(&hsv[i], 1, ch, noArray(), hist[i], 2, histSize, ranges, true) # CPP
    # mask = cv2.bitwise_xor(img, img)
    channels = [0, 1, 2]
    histSize = [8, 10, 10]
    ranges = [0, 180, 0, 256, 0, 256]
    hist = cv2.calcHist([hsv], channels, None, histSize, ranges)
    # cv2.imshow("hist", hist)
    hist = cv2.normalize(hist, hist, 1.0, 0.0, cv2.NORM_MINMAX)
    # print(hist)
    return hist


def hist2signature(hist):
    histSize = hist.shape
    assert len(histSize) == 3
    signature = numpy.zeros(shape=(histSize[0] * histSize[1] * histSize[2], 4), dtype=numpy.float32)
    for h in range(histSize[0]):
        for s in range(histSize[1]):
            for v in range(histSize[2]):
                idx = (h + 1) * (s + 1) * (v + 1) - 1
                signature[idx][0] = hist[h][s][v]
                signature[idx][1] = h
                signature[idx][2] = s
                signature[idx][3] = v
    return signature


if __name__ == '__main__':
    path_rgb = r'./video/scivideo/vid_rgb_frame1.png'
    img_rgb = cv2.imread(path_rgb)
    # print(img_rgb)
    # img_down = cv2.pyrUp(cv2.pyrDown(img_rgb))
    img_down = cv2.pyrDown(img_rgb)
    # cv2.imshow("img_rgb", img_rgb)
    # 计算直方图
    hist_gt = get_hist(img_rgb)
    hist_down = get_hist(img_down)
    # EMD算法要求的数据格式不是计算的hist，而是(val, h, s, v) * (h_bin*s_bin*v_bin)  因此需要转换
    signature_hist = hist2signature(hist_down)
    signature_gt = hist2signature(hist_gt)
    # 计算 EMD
    retval, lowerBound, flow = cv2.EMD(signature_hist, signature_gt, cv2.DIST_L2)
    print(retval)
    cv2.waitKey(0)

C++

先贴上别的大佬的实现吧

https://blog.csdn.net/qq_30815237/article/details/86807253?utm_medium=distribute.pc_relevant.none-task-blog-baidujs_title-0&spm=1001.2101.3001.4242

https://blog.csdn.net/z827997640/article/details/79841372