Python图片检索之以图搜图_Python

一、待搜索图

Python图片检索之以图搜图

二、测试集

Python图片检索之以图搜图

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)]