图像重合度算法实验

目录

一：直方图算法

二，hash算法

 三 ssim算法

---

一：直方图算法

目录

一：直方图算法

---

思想：先把两张图片切割，然后让第二张图片的切片去对比第一张图片的切片。

两张图片如下：

代码实践：image2Comparison.py

# -*- coding: utf-8 -*-
import cv2
import numpy as np
from skimage.metrics import structural_similarity as compare_ssim
import matplotlib.pyplot as plt

# 通过得到RGB每个通道的直方图来计算相似度
def classify_hist_with_split(image1, image2):
    # 将图像分离为RGB三个通道，再计算每个通道的相似值
    arr_image1 = cv2.split(image1)
    arr_image2 = cv2.split(image2)
    sub_data = 0
    npart = 10  #纵向分割
    x_arr = [i for i in range(npart)]#np.arange(npart)
    y1_arr = []
    y2_arr = []
    y3_arr = []
    wp1,wp2 = image1.shape[1]/npart,image2.shape[1]/npart
    h_img2 = arr_image2[0][:,0:round(wp2)]  #第二张图片的第一通道的第1片
    for i in range(0,npart):   #循环对比第一张图片
        h_img1 = arr_image1[0][:,round(wp1*i):round(wp1*(i+1))]
        idata = calculate(h_img2,h_img1)
        y1_arr.append(idata)
        print(idata)
    print("-----------")
    max1_index = y1_arr.index(max(y1_arr))
    h_img2 = arr_image2[1][:,0:round(wp2)]  #第二张图片的第二通道的第1片
    for i in range(0,npart):   #循环对比第一张图片
        h_img1 = arr_image1[1][:,round(wp1*i):round(wp1*(i+1))]
        idata = calculate(h_img2,h_img1)
        y2_arr.append(idata)
        print(idata)
    print("-----------")
    max2_index = y2_arr.index(max(y2_arr))
    h_img2 = arr_image2[2][:,0:round(wp2)]  #第二张图片的第三通道的第1片
    for i in range(0,npart):   #循环对比第一张图片
        h_img1 = arr_image1[2][:,round(wp1*i):round(wp1*(i+1))]
        idata = calculate(h_img2,h_img1)
        y3_arr.append(idata)
        print(idata)
    max3_index = y3_arr.index(max(y3_arr))
    #plt.plot(x_arr,y1_arr,"r:x")
    plt.plot(x_arr,y1_arr,"r:x",
             x_arr,y2_arr,"b-D",
             x_arr,y3_arr,"y--_")
    plt.savefig(r"c123.png",dpi=75)
    plt.show()
    print("-----------")
    print("%d,%d,%d" % (max1_index,max2_index,max3_index))
    sub_data = (max1_index+max2_index+max3_index+3)/(npart*3)
    return sub_data
# 计算单通道的直方图的相似值
def calculate(image1, image2):
    hist1 = cv2.calcHist([image1], [0], None, [256], [0.0, 255.0])
    hist2 = cv2.calcHist([image2], [0], None, [256], [0.0, 255.0])
    # 计算直方图的重合度
    degree = 0
    for i in range(len(hist1)):
        if hist1[i] != hist2[i]:
            degree = degree + (1 - abs(hist1[i] - hist2[i]) / max(hist1[i], hist2[i]))
        else:
            degree = degree + 1
    degree = degree / len(hist1)
    return degree
def main():
    img1 = cv2.imread(r'D:\data\real\Thumbnails\0.JPG')
    img2 = cv2.imread(r'D:\data\real\Thumbnails\1.JPG')
    n = classify_hist_with_split(img1, img2)
    print('三色直方图算法相似度：', n)


if __name__=="__main__":
    main()

运行结果(npart=10)：三色直方图算法相似度： 0.5

运行结果绘图：

 运行结果（npart=50时）：三色直方图算法相似度： 0.43333333333333335

运行结果图片：

二，hash算法

思想：分别切割两个图片（10x10）

代码（平均hash算法）：

# -*- coding: utf-8 -*-
import cv2
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

def classify_with_split(image1, image2):
    npart = 10
    #每张图片分割成10x10份
    hd1,wd1 = image1.shape[0]/npart,image1.shape[1]/npart
    hd2,wd2 = image2.shape[0]/npart,image2.shape[1]/npart
    #分别算每张割片的hash值
    hash_arr1 = np.zeros((npart,npart,100))
    hash_arr2 = np.zeros((npart,npart,100))
    for i2 in range(0,npart):
        for j2 in range(0,npart):
            p_img2 = image2[round(i2*hd2):round((i2+1)*hd2) , round(wd2*j2):round(wd2*(j2+1))]
            hash_arr2[i2,j2] = pHash(p_img2)
    for i1 in range(0,npart):
        for j1 in range(0,npart):
            p_img1 = image1[round(i1*hd1):round((i1+1)*hd1) , round(wd1*j1):round(wd1*(j1+1))]
            hash_arr = pHash(p_img1)
            print("[%d,%d]=%s" % (i1,j1,hash_arr))
            hash_arr1[i1,j1] = hash_arr
    #第二张图片的第1张割片，去匹配第一张图片的所有割片，并创建热力图
    match_arr =  np.zeros((npart,npart))
    # for x in np.nditer(hash_arr1):
    #     match_arr.append(cmpHash(x,hash_arr2[0][0]))
    # match_arr = match_arr.reshape(npart,npart)
    for i in range(len(hash_arr1)):
        for j in range(len(hash_arr1[i])):
            match_arr[i,j]=cmpHash(hash_arr1[i][j],hash_arr2[0][0])
    print(match_arr)
    #绘图
    f, ax = plt.subplots(figsize=(10, 10))
    sns.heatmap(match_arr, ax=ax,cmap='YlOrRd',linewidths=0.1,linecolor="grey",cbar_kws={"orientation":"horizontal"})

    plt.savefig(r"hash.png",dpi=75)
    plt.show()
    pos_max = np.unravel_index(np.argmax(match_arr),match_arr.shape)
    rate = (npart-pos_max[0])*(npart-pos_max[1])/(npart * npart)
    return rate

# Hash值对比
def cmpHash(hash1, hash2,shape=(10,10)):
    n = 0
    # hash长度不同则返回-1代表传参出错
    if len(hash1)!=len(hash2):
        return -1
    # 遍历判断
    for i in range(len(hash1)):
        # 相等则n计数+1，n最终为相似度
        if hash1[i] == hash2[i]:
            n = n + 1
    return n/(shape[0]*shape[1])

def main():
    img1 = cv2.imread(r'D:\data\real\Thumbnails\0.JPG')
    img2 = cv2.imread(r'D:\data\real\Thumbnails\1.JPG')
    n = classify_with_split(img1, img2)
    print('hash算法相似度：', n)


if __name__=="__main__":
    main()

运行结果：hash算法相似度： 0.45

运行图片：

 三 ssim算法

思想：同上

代码：

import cv2
import numpy as np
from skimage.metrics import structural_similarity as compare_ssim
import matplotlib.pyplot as plt
import seaborn as sns

def classify_with_split(image1, image2):
    npart = 10
    #每张图片分割成10x10份
    hd1,wd1 = image1.shape[0]/npart,image1.shape[1]/npart
    hd2,wd2 = image2.shape[0]/npart,image2.shape[1]/npart
    #分别算每张割片的hash值
    hash_arr1 = np.zeros((npart,npart,100))
    hash_arr2 = np.zeros((npart,npart,100))
    #第二张图片的第1张割片，去匹配第一张图片的所有割片，并创建热力图
    match_arr =  np.zeros((npart,npart))
    for i1 in range(0,npart):
        for j1 in range(0,npart):
            p_img1 = image1[round(i1*hd1):round((i1+1)*hd1) , round(wd1*j1):round(wd1*(j1+1))]
            p_img2 = image2[0:round(hd2) , 0:round(wd2)]
            ssimRate = compare_ssim(p_img1, p_img2, multichannel=True)
            match_arr[i1,j1]=ssimRate
            print("[%d,%d]vs[%d,%d]=%.2f"%(0,0,i1,j1,ssimRate))
    print(match_arr)
    #绘图
    f, ax = plt.subplots(figsize=(10, 10))
    sns.heatmap(match_arr, ax=ax,cmap='YlOrRd',linewidths=0.1,linecolor="grey",cbar_kws={"orientation":"horizontal"})

    plt.savefig(r"ssim.png",dpi=75)
    plt.show()
    pos_max = np.unravel_index(np.argmax(match_arr),match_arr.shape)
    print(pos_max)
    rate = (npart-pos_max[0])*(npart-pos_max[1])/(npart * npart)
    return rate

def main():
    img1 = cv2.imread(r'D:\data\real\Thumbnails\0.JPG')
    img2 = cv2.imread(r'D:\data\real\Thumbnails\1.JPG')
    n = classify_with_split(img1, img2)
    print('ssim算法相似度：', n)


if __name__=="__main__":
    main()

运行结果：hash算法相似度： 0.8

运行结果图片：

四：PSNR算法

代码：

def PSNR(img1, img2):
    mse = np.mean((img1/255. - img2/255.) ** 2)
    if mse == 0:
        return 100
    PIXEL_MAX = 1
    return 20 * math.log10(PIXEL_MAX / math.sqrt(mse))

运行结果：

(0, 2)

目录

一：直方图算法

二，hash算法

 三 ssim算法

四：PSNR算法

---

hash算法相似度： 0.8

运行结果图：