Python圖片檢索之以圖搜圖

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

from image_similarity_function import *
import os
import shutil

# 融合相似度閾值
threshold1 = 0.70
# 最終相似度較高判斷閾值
threshold2 = 0.95


# 融合函數(shù)計(jì)算圖片相似度
def calc_image_similarity(img1_path, img2_path):
    """
    :param img1_path: filepath+filename
    :param img2_path: filepath+filename
    :return: 圖片最終相似度
    """

    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))
    # 如果三種算法的相似度最大的那個(gè)大于0.7，則相似度取最大，否則，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__':

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

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

    # 相似圖片存放路徑
    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
                    # 比較
                    kk = calc_image_similarity(img1_path, img2_path)
                    try:
                        if kk >= threshold2:
                            # 將兩張照片同時(shí)拷貝到指定目錄
                            shutil.copy(img2_path, os.path.join(newfilepath, srcfilename[:-4] + "_" + filename))
                    except Exception as e:
                        # print(e)
                        pass

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

# 導(dǎo)入包
import cv2
from functools import reduce
from PIL import Image


# 計(jì)算兩個(gè)圖片相似度函數(shù)ORB算法
def ORB_img_similarity(img1_path, img2_path):
    """
    :param img1_path: 圖片1路徑
    :param img2_path: 圖片2路徑
    :return: 圖片相似度
    """
    try:
        # 讀取圖片
        img1 = cv2.imread(img1_path, cv2.IMREAD_GRAYSCALE)
        img2 = cv2.imread(img2_path, cv2.IMREAD_GRAYSCALE)

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

        # 提取并計(jì)算特征點(diǎn)
        bf = cv2.BFMatcher(cv2.NORM_HAMMING)
        # knn篩選結(jié)果
        matches = bf.knnMatch(des1, trainDescriptors=des2, k=2)

        # 查看最大匹配點(diǎn)數(shù)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'


# 計(jì)算圖片的局部哈希值--pHash
def phash(img):
    """
    :param img: 圖片
    :return: 返回圖片的局部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


# 計(jì)算兩個(gè)圖片相似度函數(shù)局部敏感哈希算法
def phash_img_similarity(img1_path, img2_path):
    """
    :param img1_path: 圖片1路徑
    :param img2_path: 圖片2路徑
    :return: 圖片相似度
    """
    # 讀取圖片
    img1 = Image.open(img1_path)
    img2 = Image.open(img2_path)

    # 計(jì)算漢明距離
    distance = bin(phash(img1) ^ phash(img2)).count('1')
    similary = 1 - distance / max(len(bin(phash(img1))), len(bin(phash(img1))))
    return similary


# 直方圖計(jì)算圖片相似度算法
def make_regalur_image(img, size=(256, 256)):
    """我們有必要把所有的圖片都統(tǒng)一到特別的規(guī)格，在這里我選擇是的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)]