python 删除相似的图片_python比较删除两个相同的图片-CSDN博客

本文链接：https://blog.csdn.net/qq_36079986/article/details/128580028

文章介绍了一种使用Python处理视频裁剪生成的大量图片的方法，通过计算图片之间的SSIM（结构相似度）来判断并删除相似图片，优化数据集。同时，利用多进程提高处理效率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

python 删除相似的图片

主要应用在制作数据集时，将视频裁剪为图片，但多少帧取一张图片不太好选择，太密导致裁剪下来的图片太多，有很多重复的；太稀疏会丢失部分有用信息。可以选择在采集时密集一些，之后使用图片相似度判断，删除一些帧，以下是代码。

# coding: utf-8
import os
import cv2
# from skimage.measure import compare_ssim
# from skimage.metrics import _structural_similarity
from skimage.metrics import structural_similarity as ssim
# import multiprocessing
import multiprocessing
import time
# import Queue
from tkinter import _flatten
def delete(filename1):
    os.remove(filename1)


def list_all_files(root):
    files = []
    list = os.listdir(root)

    if len(list)>0:
        if (not os.path.isdir(os.path.join(root, list[0]))) and list[0].endswith('.jpg'):  # os.path.isdir()用于判断某一对象(需提供绝对路径)是否为目录
            dict = {}
            for l in list:
                l_ = l.split('.')[0]

                dict[int(l_.split('_')[-2])*1000000 + int(l_.split('_')[-1])] = l
            dict_keys = sorted(dict.keys())
            list = []
            print(dict_keys)

            for dict_key in dict_keys:
                list.append(dict[dict_key])
            # print('list',list)

    # os.listdir()方法：返回指定文件夹包含的文件或子文件夹名字的列表。该列表顺序以字母排序
    for i in range(len(list)):
        element = os.path.join(root, list[i])
        # 需要先使用python路径拼接os.path.join()函数，将os.listdir()返回的名称拼接成文件或目录的绝对路径再传入os.path.isdir()和os.path.isfile().
        if os.path.isdir(element):  # os.path.isdir()用于判断某一对象(需提供绝对路径)是否为目录
            # temp_dir = os.path.split(element)[-1]
            # os.path.split分割文件名与路径,分割为data_dir和此路径下的文件名，[-1]表示只取data_dir下的文件名
            files.append(list_all_files(element))

        elif os.path.isfile(element) and type(element)==str:
            files.append(element)
        else:
            print('err',element)
            raise element
    print('2',files)
    return files



def simJr(count,left_index, right_index,img_files,q,interval):
    # print('llllllllllllllllll')
    # left_index,right_index = q.get()
    print('77img_files,left_index,',left_index,img_files[left_index])
    img = cv2.imread(img_files[left_index])
    img1 = cv2.imread(img_files[right_index])
    # print('76big_ssim:', left_index, right_index, img_files[left_index], img_files[right_index])
    # 进行结构性相似度判断
    # ssim_value = _structural_similarity.structural_similarity(img,img1,multichannel=True)
    ssim_value = ssim(img, img1, multichannel=True)
    # ssim_value = pool.apply_async(func=ssim, args=(img, img1,True,))
    print('67big_ssim:', left_index, right_index, img_files[left_index], img_files[right_index], ssim_value)
    if ssim_value > 0.5:
        # print('big_ssim:', left_index, right_index, img_files[left_index], img_files[right_index], ssim_value)
        delete(img_files[right_index])
        print('86delete', right_index, img_files[right_index])
        right_index += 1
    else:
        left_index = right_index
        right_index += interval
    q.put([left_index,right_index])

def q_clear(q):
    while q.qsize()>0:
        q.get()


if __name__ == '__main__':
    pool = multiprocessing.Pool(processes=3)
    img_path = r' '

    q = multiprocessing.Queue()
    count = 0
    left_index = 0
    right_index = 1
    count = 0
    # interval=2
    P_num = 10
    for interval in range(1,50):
        try:
            imgs_n = []
            all_files = list_all_files(path)  # 返回包含完整路径的所有图片名的列表
            all_files = list(_flatten(all_files))
            print('1', len(all_files), all_files)

            img_files = all_files

            q_clear(q)
            for i in range(P_num):
                print('[int((len(all_files) / P_num) * i), int((len(all_files) / P_num) * i) + 1]',[int((len(all_files) / P_num) * i), int((len(all_files) / P_num) * i) + interval])
                q.put([int((len(all_files) / P_num) * i), int((len(all_files) / P_num) * i) + interval])

            flag_begin = True
            while right_index < len(all_files) or flag_begin == True:
                left_index, right_index = q.get()
                # pool.apply_async(func=simJr, args=(count,left_index, right_index,img_files,q,))
                simJr_P  = multiprocessing.Process(target=simJr, args=(count,left_index, right_index,img_files,q,interval,))
                simJr_P.start()

                print('q.qsize()',q.qsize(),'count',count)
                print('122left_index, right_index',left_index, right_index)
                count+=1
                flag_begin = False
        except:
            print('err_overinterval',interval)
            time.sleep(1000)
            q_clear(q)