#!/usr/bin/env python # -*- coding: utf-8 -*- # @Time : 2024/5/24 0024 上午 10:09 # @Author : liudan # @File : image_similarity_count.py # @Software: pycharm import cv2 import os from PIL import Image, ImageDraw import numpy as np from skimage.metrics import structural_similarity as compare_ssim import json import random import yaml import demo_env from demo_env import registration_demo from sklearn.metrics.pairwise import cosine_similarity def compare_boxes_similarity(image1_path, image2_path, json_file_path, similarity_threshold=0.4): try: if not os.path.exists(image1_path): raise FileNotFoundError(f"Image file {image1_path} not found") image1 = cv2.imread(image1_path) # image1 = np.array(Image.open(image1_path)) # 原图尺寸，未resize # image1 = image1[:, :, ::-1] image2 = wrap_image # draw1 = ImageDraw.Draw(image1) # 存储相似度结果和是否相同的判断 similarity_results = [] same_content_boxes = [] with open(json_file_path, 'r') as f: data = json.load(f) for shape in data['shapes']: if 'points' in shape: shape['points'] = [[int(round(x)), int(round(y))] for x, y in shape['points']] x1, y1 = shape['points'][0] x2, y2 = shape['points'][1] # 从两幅图像中截取对应区域 # region1 = image1.crop((x1, y1, x2, y2)) # # region1 = image1.crop((x1, y1, x2, y2)).convert('L') # draw1.rectangle([x1, y1, x2, y2], outline='red', width=2) # image1.save(os.path.join(params['save_dir'], f'save_annotated1_{i}.jpg')) # region1.save(os.path.join(params['save_dir'], f'111111{i}.jpg')) # region1 = image1[y1:y2, x1:x2] filename = f'json_image1_{shape["label"]}_{i}.jpg' cv2.imwrite(os.path.join(params['save_dir'], filename),region1) cv2.rectangle(image1, (x1, y1), (x2, y2), (0, 255, 0), 2) cv2.imwrite(os.path.join(params['save_dir'], f'save_annotated1_{i}.jpg'), image2) # region2 = image2.crop((left-80, top, right-80, bottom)) region2 = image2[y1:y2, x1:x2] # region2 = cv2.cvtColor(region2, cv2.COLOR_BGR2GRAY) # region2= region2.transpose(Image.FLIP_TOP_BOTTOM) #旋转180°针对pillowImage对象 # region2 = cv2.rotate(region2, cv2.ROTATE_180) filename = f'json_image2_{shape["label"]}_{i}.jpg' cv2.imwrite(os.path.join(params['save_dir'], filename), region2) cv2.rectangle(image2, (x1, y1), (x2, y2),(0,255,0), 2) cv2.imwrite(os.path.join(params['save_dir'], f'save_annotated2_{i}.jpg'), image2) # 将PIL图像转换为numpy数组，以便进行计算 arr1 = np.array(region1) arr2 = region2 # region2一直是numpy数组，所以上述image1和image2处理方式不同 # 确保两个数组的形状是相同的 assert arr1.shape == arr2.shape, "Images do not have the same size for the given box" # 使用SSIM计算相似度（范围在-1到1之间，1表示完全相似） # ssim = compare_ssim(arr1, arr2, multichannel=False) # 这是旧版，可以计算灰度图相似度，对于计算彩色图像即使设置multichannel=True也错 # ssim = compare_ssim(arr1, arr2, channel_axis=2) ssim = batch_ssim(arr1, arr2) similarity_results.append(ssim) if ssim > similarity_threshold: same_content_boxes.append(shape) cv2.rectangle(image2, (x1, y1), (x2, y2),(0,255,0), 2) text = "score: " + str(round(ssim, 3)) text_pos = (x1, y1 - 5) # 参数：图像, 文本, 文本位置, 字体类型, 字体大小, 字体颜色, 字体粗细 cv2.putText(image2, text, text_pos, cv2.FONT_HERSHEY_SIMPLEX, 1.5, (0, 255, 0), 2) cv2.imwrite(os.path.join(params['visualization_dir'],f'{wrap_images_name[:-8]}_{i}.jpg'), image2) else: cv2.rectangle(image2, (x1, y1), (x2, y2), (0, 0, 255), 2) text = "score: " + str(round(ssim, 3)) text_pos = (x1, y1 - 5) # 参数：图像, 文本, 文本位置, 字体类型, 字体大小, 字体颜色, 字体粗细 cv2.putText(image2, text, text_pos, cv2.FONT_HERSHEY_SIMPLEX, 1.5, (0, 0, 255), 2) cv2.imwrite(os.path.join(params['visualization_dir'], f'{wrap_images_name[:-8]}_{i}.jpg'), image2) return similarity_results, same_content_boxes except FileNotFoundError as e: print(f"An error occurred: {e}") except Exception as e: print(f"An unexpected error occurred: {e}") return None, None def read_params_from_yml(yml_file_path): with open(yml_file_path, 'r') as file: params = yaml.safe_load(file) return params def batch_ssim(im1, im2): imgsize = im1.shape[1] * im1.shape[2] avg1 = im1.mean((1, 2), keepdims=1) avg2 = im2.mean((1, 2), keepdims=1) std1 = im1.std((1, 2), ddof=1) std2 = im2.std((1, 2), ddof=1) cov = ((im1 - avg1) * (im2 - avg2)).mean((1, 2)) * imgsize / (imgsize - 1) avg1 = np.squeeze(avg1) avg2 = np.squeeze(avg2) k1 = 0.01 k2 = 0.03 c1 = (k1 * 255) ** 2 c2 = (k2 * 255) ** 2 c3 = c2 / 2 # return np.mean((cov + c3) / (std1 * std2 + c3)) return np.mean( (2 * avg1 * avg2 + c1) * 2 * (cov + c3) / (avg1 ** 2 + avg2 ** 2 + c1) / (std1 ** 2 + std2 ** 2 + c2)) if __name__ == "__main__": yml_file_path = 'params.yml' params = read_params_from_yml(yml_file_path) wrap_images_all = registration_demo(params['image_dir'],params['demo_image_path'], params['json_ref_path'], params['ref_image_path']) for i, item in enumerate(wrap_images_all): wrap_image,wrap_images_name = item similarity_results, same_content_boxes = compare_boxes_similarity(params['path_to_image1'], wrap_image, params['json_file_path'], params['similarity_threshold']) # 打印所有坐标框的相似度结果 print(f"{wrap_images_name}\n") for idx, score in enumerate(similarity_results, 1): print(f"Similarity Score for Box {idx}: {score}") # 打印被认为是相同内容的坐标框 print("Boxes with the same content:") for shape in same_content_boxes: print(shape['label'] + ' object is same as template')