Python图片检索之以图搜图

一、待搜索图

二、测试集

三、new_similarity_compare.py

# -*- encoding=utf-8 -*-

from image_similarity_function import *
import os
import shutil

# 融合相似度阈值
threshold1 = 0.70
# 最终相似度较高判断阈值
threshold2 = 0.95

# 融合函数计算图片相似度
def calc_image_similarity(img1_path, img2_path):
   """
   :param img1_path: filepath+filename
   :param img2_path: filepath+filename
   :return: 图片最终相似度
   """

similary_ORB = float(ORB_img_similarity(img1_path, img2_path))
   similary_phash = float(phash_img_similarity(img1_path, img2_path))
   similary_hist = float(calc_similar_by_path(img1_path, img2_path))
   # 如果三种算法的相似度最大的那个大于0.7，则相似度取最大，否则，取最小。
   max_three_similarity = max(similary_ORB, similary_phash, similary_hist)
   min_three_similarity = min(similary_ORB, similary_phash, similary_hist)
   if max_three_similarity > threshold1:
       result = max_three_similarity
   else:
       result = min_three_similarity

return round(result, 3)

if __name__ == '__main__':

# 搜索文件夹
   filepath = r'D:\Dataset\cityscapes\leftImg8bit\val\frankfurt'

#待查找文件夹
   searchpath = r'C:\Users\Administrator\Desktop\cityscapes_paper'

# 相似图片存放路径
   newfilepath = r'C:\Users\Administrator\Desktop\result'

for parent, dirnames, filenames in os.walk(searchpath):
       for srcfilename in filenames:
           img1_path = searchpath +"\\"+ srcfilename
           for parent, dirnames, filenames in os.walk(filepath):
               for i, filename in enumerate(filenames):
                   print("{}/{}: {} , {} ".format(i+1, len(filenames), srcfilename,filename))
                   img2_path = filepath + "\\" + filename
                   # 比较
                   kk = calc_image_similarity(img1_path, img2_path)
                   try:
                       if kk >= threshold2:
                           # 将两张照片同时拷贝到指定目录
                           shutil.copy(img2_path, os.path.join(newfilepath, srcfilename[:-4] + "_" + filename))
                   except Exception as e:
                       # print(e)
                       pass

四、image_similarity_function.py

# -*- encoding=utf-8 -*-

# 导入包
import cv2
from functools import reduce
from PIL import Image

# 计算两个图片相似度函数ORB算法
def ORB_img_similarity(img1_path, img2_path):
   """
   :param img1_path: 图片1路径
   :param img2_path: 图片2路径
   :return: 图片相似度
   """
   try:
       # 读取图片
       img1 = cv2.imread(img1_path, cv2.IMREAD_GRAYSCALE)
       img2 = cv2.imread(img2_path, cv2.IMREAD_GRAYSCALE)

# 初始化ORB检测器
       orb = cv2.ORB_create()
       kp1, des1 = orb.detectAndCompute(img1, None)
       kp2, des2 = orb.detectAndCompute(img2, None)

# 提取并计算特征点
       bf = cv2.BFMatcher(cv2.NORM_HAMMING)
       # knn筛选结果
       matches = bf.knnMatch(des1, trainDescriptors=des2, k=2)

# 查看最大匹配点数目
       good = [m for (m, n) in matches if m.distance < 0.75 * n.distance]
       similary = len(good) / len(matches)
       return similary

except:
       return '0'

# 计算图片的局部哈希值--pHash
def phash(img):
   """
   :param img: 图片
   :return: 返回图片的局部hash值
   """
   img = img.resize((8, 8), Image.ANTIALIAS).convert('L')
   avg = reduce(lambda x, y: x + y, img.getdata()) / 64.
   hash_value = reduce(lambda x, y: x | (y[1] << y[0]), enumerate(map(lambda i: 0 if i < avg else 1, img.getdata())),
                       0)
   return hash_value

# 计算两个图片相似度函数局部敏感哈希算法
def phash_img_similarity(img1_path, img2_path):
   """
   :param img1_path: 图片1路径
   :param img2_path: 图片2路径
   :return: 图片相似度
   """
   # 读取图片
   img1 = Image.open(img1_path)
   img2 = Image.open(img2_path)

# 计算汉明距离
   distance = bin(phash(img1) ^ phash(img2)).count('1')
   similary = 1 - distance / max(len(bin(phash(img1))), len(bin(phash(img1))))
   return similary

# 直方图计算图片相似度算法
def make_regalur_image(img, size=(256, 256)):
   """我们有必要把所有的图片都统一到特别的规格，在这里我选择是的256x256的分辨率。"""
   return img.resize(size).convert('RGB')

def hist_similar(lh, rh):
   assert len(lh) == len(rh)
   return sum(1 - (0 if l == r else float(abs(l - r)) / max(l, r)) for l, r in zip(lh, rh)) / len(lh)

def calc_similar(li, ri):
   return sum(hist_similar(l.histogram(), r.histogram()) for l, r in zip(split_image(li), split_image(ri))) / 16.0

def calc_similar_by_path(lf, rf):
   li, ri = make_regalur_image(Image.open(lf)), make_regalur_image(Image.open(rf))
   return calc_similar(li, ri)

def split_image(img, part_size=(64, 64)):
   w, h = img.size
   pw, ph = part_size
   assert w ％ pw == h ％ ph == 0
   return [img.crop((i, j, i + pw, j + ph)).copy() for i in range(0, w, pw) \
           for j in range(0, h, ph)]

五、结果

来源：https://blog.csdn.net/weixin_43723625/article/details/117298412