python图片相似度识别算法_图片相似度识别：pHash算法

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

import pandas as pd

import cv2

import time

import numpy as np

def pHash(img,leng=32,wid=32):

img = cv2.resize(img, (leng, wid))

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

dct = cv2.dct(np.float32(gray))

dct_roi = dct[0:8, 0:8]

avreage = np.mean(dct_roi)

phash_01 = (dct_roi>avreage)+0

phash_list = phash_01.reshape(1,-1)[0].tolist()

hash = ''.join([str(x) for x in phash_list])

return hash

def dHash(img,leng=9,wid=8):

img=cv2.resize(img,(leng, wid))

image=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

#每行前一个像素大于后一个像素为1，相反为0，生成哈希

hash=[]

for i in range(wid):

for j in range(wid):

if image[i,j]>image[i,j+1]:

hash.append(1)

else:

hash.append(0)

return hash

def aHash(img,leng=8,wid=8):

img=cv2.resize(img,(leng, wid))

image=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

avreage = np.mean(image)

hash = []

for i in range(image.shape[0]):

for j in range(image.shape[1]):

if image[i,j] >= avreage:

hash.append(1)

else:

hash.append(0)

return hash

def Hamming_distance(hash1,hash2):

num = 0

for index in range(len(hash1)):

if hash1[index] != hash2[index]:

num += 1

return num

if __name__ == '__main__':

image1 = cv2.imread('image1')

image2 = cv2.imread('image2')

start1 = time.time()

d_dist = Hamming_distance(dHash(image1),dHash(image2))

end1 = time.time()

start2 = time.time()

p_dist = Hamming_distance(pHash(image1),pHash(image2))

end2 = time.time()

start3 = time.time()

a_dist = Hamming_distance(aHash(image1),aHash(image2))

end3 = time.time()

print('a_dist is '+'%d' % a_dist + ', similarity is ' +'%f' % (1 - a_dist * 1.0 / 64) + ', time is ' +'%f' % (end3-start3))

print('p_dist is '+'%d' % p_dist + ', similarity is ' +'%f' % (1 - p_dist * 1.0 / 64) + ', time is ' +'%f' % (end2-start2))

print('d_dist is '+'%d' % d_dist + ', similarity is ' +'%f' % (1 - d_dist * 1.0 / 64) + ', time is ' +'%f' % (end1-start1))