图像相似度计算

利用直方图特征计算图像之间的相似度，得到相关矩阵

# -*- coding: utf-8 -*-

import cv2
import os
import math
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np

#获取图片地址
def get_imlist(path):
    return [os.path.join(path,f) for f in os.listdir(path) if f.endswith('.jpg')]

def transferimgsize(imlist):
    #转换图片成相同尺寸
    for img in imlist:
        pic=cv2.imread(img)
        pic = cv2.resize(pic, (200, 80), interpolation=cv2.INTER_CUBIC)
        cv2.imwrite(img, pic)


def get_hsv_hist_feature(image_name):
    # 获取图像hsv直方图特征，返回h,s,v三个维度各自的直方图
    img = cv2.imread(image_name)
    hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
    arr_h=hsv[:,:,0].flatten()
    arr_s=hsv[:,:,1].flatten()
    arr_v=hsv[:,:,2].flatten()
    nh, bins, patches= plt.hist(arr_h, bins=256, normed=1)
    ns, bins, patches = plt.hist(arr_s, bins=256, normed=1)
    nv, bins, patches = plt.hist(arr_v, bins=256, normed=1)
    return [nh,ns,nv]

def cal_his_intersection(image_name1,image_name2):
    # 计算2张图片的颜色特征的相似度
    value=0
    feature1=get_hsv_hist_feature(image_name1)
    feature2=get_hsv_hist_feature(image_name2)
    for i in range(2):
        count1=feature1[i]
        count2=feature2[i]
        Sum1=sum(count1);Sum2=sum(count2)
        Sumup = [math.sqrt(a*b) for a,b in zip(count1,count2)]
        SumDown = math.sqrt(Sum1*Sum2)
        Sumup = sum(Sumup)
        HistDist=1-math.sqrt(1-Sumup/SumDown)
        value+=HistDist
    return value/3

def cal_Euclidean_Distance(image_name1,image_name2):
    # 计算2张图片的形状、质感等特征的相似度
    img1 = cv2.imread(image_name1)
    img2 = cv2.imread(image_name2)
    img1 = cv2.cvtColor(img1, cv2.COLOR_RGB2GRAY)
    img2 = cv2.cvtColor(img2, cv2.COLOR_RGB2GRAY)
    dist=np.linalg.norm(img1-img2)
    return dist

if __name__ == "__main__":
    imlist = get_imlist('D:/mywork/python/Images/image66')
    final=pd.DataFrame(data=None,index=imlist, columns=imlist)
    for img_i in imlist:
        for img_j in imlist:
            #记录相似度矩阵数据
            final.loc[img_i,img_j]=cal_Euclidean_Distance(img_i,img_j)