删除重复图像

import os
import cv2
import numpy as np
from keras.applications.resnet50 import ResNet50, preprocess_input
 
# pip install opencv-python numpy keras tensorflow
 
def extract_image_features(image_path):
    image = cv2.imread(image_path)  # 读取图片
    image = cv2.resize(image, (256, 256))  # 缩放图片到统一尺寸
    image = image[16:240, 16:240]  # 裁剪中间区域(224x224)
    
    image = np.expand_dims(image, axis=0)  # 扩展维度以匹配模型输入要求
    image = preprocess_input(image)  # 预处理图片
    
    features = model.predict(image)  # 提取特征向量
    features /= np.linalg.norm(features)  # 归一化特征向量
    
    return features.flatten()  # 平铺特征向量
 
def delete_duplicate_images():
    current_dir = os.getcwd()  # 获取当前目录路径
    files = [f for f in os.listdir(current_dir) if os.path.isfile(os.path.join(current_dir, f))]  # 获取当前目录下的所有文件
 
    image_features = {}
    deleted_count = 0  # 记录删除的图片数量
    duplicate_pairs = []  # 用于保存重复图片的文件名对
 
    for file_name in files:
        if file_name.endswith(".jpg") or file_name.endswith(".png"):  # 筛选出图片文件
            file_path = os.path.join(current_dir, file_name)
            image_feature = extract_image_features(file_path)
 
            is_duplicate = False
            for existing_path, existing_feature in image_features.items():
                distance = np.linalg.norm(existing_feature - image_feature)  # 计算欧氏距离
                if distance < 0.3:  # 设定阈值来判断相似度，根据实际情况调整
                    is_duplicate = True
                    print(f"删除重复图片: {file_path}")
                    os.remove(file_path)
                    deleted_count += 1
                    # 记录重复的文件名对 (当前文件名和已有文件名)
                    duplicate_pairs.append((file_name, os.path.basename(existing_path)))
                    break
 
            if not is_duplicate:
                image_features[file_path] = image_feature
 
    # 将重复图片文件名对保存到txt文件
    if duplicate_pairs:
        with open("duplicate_images.txt", "w") as f:
            for file1, file2 in duplicate_pairs:
                f.write(f"{file1} 与 {file2} 重复\n")
    
    print("已删除 {} 张重复图片".format(deleted_count))
 
# 加载预训练的ResNet50模型
model = ResNet50(weights='imagenet', include_top=False, pooling='avg')
 
delete_duplicate_images()