import numpy as np
import torch
import cv2
import matplotlib.pyplot as plt
import os
from pathlib import Pathdef read_raw_gt(file_path, height, width, bl):"""读取原始RAW文件并进行预处理"""raw_image = np.fromfile(file_path, dtype=np.int16)raw_image = raw_image.reshape((height, width)).astype(np.float32)raw_image = np.maximum(raw_image - bl, 0) / (4095 - bl) # 归一化到[0,1]raw_image = np.expand_dims(raw_image, axis=2) # 增加通道维度# 分割为RGGB四通道raw_out = np.concatenate((raw_image[0:height:2, 0:width:2, :], # Rraw_image[0:height:2, 1:width:2, :], # Graw_image[1:height:2, 0:width:2, :], # Graw_image[1:height:2, 1:width:2, :]), axis=2)return raw_out # 形状 (H//2, W//2, 4)def RGGB_t_2img(raw_tensor):"""将四通道RGGB tensor转换为RGB图像"""raw_image = raw_tensor.numpy() if torch.is_tensor(raw_tensor) else raw_tensorchannels, height, width = raw_image.shapeassert channels == 4, "Tensor must have 4 channels"# 构建拜耳阵列bayer = np.zeros((height * 2, width * 2), dtype=np.uint8)bayer[0::2, 0::2] = (raw_image[0] * 255).astype(np.uint8) # Rbayer[0::2, 1::2] = (raw_image[1] * 255).astype(np.uint8) # Gbayer[1::2, 0::2] = (raw_image[2] * 255).astype(np.uint8) # Gbayer[1::2, 1::2] = (raw_image[3] * 255).astype(np.uint8) # B# 解拜耳转换为RGBrgb = cv2.cvtColor(bayer, cv2.COLOR_BayerRGGB2RGB)return rgbdef calculate_laplacian_variance(rgb_image):"""计算拉普拉斯方差清晰度指标"""gray = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2GRAY)laplacian = cv2.Laplacian(gray, cv2.CV_64F)return laplacian.var()def generate_blur_heatmap(rgb_image, window_size=64):"""生成局部模糊热力图"""gray = cv2.cvtColor(rgb_image, cv2.COLOR_RGB2GRAY)h, w = gray.shapeheatmap = np.zeros((h // window_size, w // window_size))for i in range(0, h // window_size):for j in range(0, w // window_size):y = i * window_sizex = j * window_sizepatch = gray[y:y + window_size, x:x + window_size]lap_var = cv2.Laplacian(patch, cv2.CV_64F).var()heatmap[i, j] = lap_var# 归一化处理heatmap = (heatmap - heatmap.min()) / (heatmap.max() - heatmap.min() + 1e-8)return cv2.resize(heatmap, (w, h), interpolation=cv2.INTER_NEAREST)def analyze_gt_blur(file_path, height, width, bl, output_dir='./results'):"""增强版分析函数，支持结果分文件保存"""# 创建输出目录Path(output_dir).mkdir(parents=True, exist_ok=True)# 生成基础文件名base_name = os.path.splitext(os.path.basename(file_path))[0]safe_name = base_name.replace(' ', '_') # 处理特殊字符# 数据处理流程（保持原逻辑）raw_out = read_raw_gt(file_path, height, width, bl)raw_tensor = torch.from_numpy(raw_out).permute(2, 0, 1)rgb_image = RGGB_t_2img(raw_tensor)global_var = calculate_laplacian_variance(rgb_image)heatmap = generate_blur_heatmap(rgb_image)# 独立保存RGB图像rgb_path = os.path.join(output_dir, f"{safe_name}_rgb.png")plt.imsave(rgb_path, rgb_image)# 独立保存热力图（含colorbar）plt.figure(figsize=(8, 6))plt.imshow(heatmap, cmap='jet')plt.colorbar()heatmap_path = os.path.join(output_dir, f"{safe_name}_heatmap.png")plt.savefig(heatmap_path, bbox_inches='tight')plt.close()# 保存组合分析图analysis_path = os.path.join(output_dir, f"{safe_name}_analysis.png")fig = plt.figure(figsize=(15, 6))# RGB图像子图ax1 = fig.add_subplot(1, 3, 1)ax1.imshow(rgb_image)ax1.set_title("RGB Image")# 热力图层ax2 = fig.add_subplot(1, 3, 2)im = ax2.imshow(heatmap, cmap='jet')fig.colorbar(im, ax=ax2)ax2.set_title("Blur Heatmap")# 直方图ax3 = fig.add_subplot(1, 3, 3)ax3.hist(heatmap.flatten(), bins=50)ax3.set_title("Blur Distribution")# 主标题fig.suptitle(f"File: {base_name}\nGlobal Laplacian Variance: {global_var:.2f}")plt.savefig(analysis_path, dpi=300, bbox_inches='tight')plt.close()return {"rgb_path": rgb_path,"heatmap_path": heatmap_path,"analysis_path": analysis_path,"global_var": global_var}import cv2
import numpy as np
from matplotlib import cmdef create_overlay(rgb_image, heatmap, alpha=0.5):"""创建热力图叠加可视化"""# 标准化热力图数据norm_heat = (heatmap - np.min(heatmap)) / (np.max(heatmap) - np.min(heatmap) + 1e-8)# 生成伪彩色热力图heatmap_color = cm.jet(norm_heat)[:, :, :3] # 转换为RGBheatmap_color = (heatmap_color * 255).astype(np.uint8)# 调整原始图像尺寸（如果需要）if rgb_image.shape[:2] != heatmap.shape[:2]:rgb_image = cv2.resize(rgb_image, (heatmap.shape[1], heatmap.shape[0]))# 创建叠加效果overlay = cv2.addWeighted(rgb_image, 1 - alpha, heatmap_color, alpha, 0)return overlaydef enhanced_analysis(file_path, height, width, bl, output_dir='./results'):"""增强版分析函数，包含叠加图生成"""# 基础处理（保持原有流程）raw_out = read_raw_gt(file_path, height, width, bl)raw_tensor = torch.from_numpy(raw_out).permute(2, 0, 1)rgb_image = RGGB_t_2img(raw_tensor)heatmap = generate_blur_heatmap(rgb_image)# 创建输出路径base_name = Path(file_path).stemsafe_name = base_name.replace(' ', '_')Path(output_dir).mkdir(exist_ok=True)# 保存原始RGBrgb_path = f"{output_dir}/{safe_name}_original.png"cv2.imwrite(rgb_path, cv2.cvtColor(rgb_image, cv2.COLOR_RGB2BGR))# 保存独立热力图plt.figure(figsize=(8, 6))plt.imshow(heatmap, cmap='jet')plt.colorbar()heatmap_path = f"{output_dir}/{safe_name}_heatmap.png"plt.savefig(heatmap_path, bbox_inches='tight')plt.close()# 生成并保存叠加图overlay = create_overlay(rgb_image, heatmap)overlay_path = f"{output_dir}/{safe_name}_overlay.png"cv2.imwrite(overlay_path, cv2.cvtColor(overlay, cv2.COLOR_RGB2BGR))return {'original': rgb_path,'heatmap': heatmap_path,'overlay': overlay_path}# 测试配置
height = 1440 # RAW文件原始高度
width = 2570 # RAW文件原始宽度height = 1080 # RAW文件原始高度
width = 1920 # RAW文件原始宽度bl = 0 # 黑电平值# 测试文件列表
raw_files = ['h-29x_200.raw','h-218x_200.raw','h-259x_200.raw'
]import osgt_dir = 'gt'
# 检查目录是否存在
if os.path.exists(gt_dir) and os.path.isdir(gt_dir):raw_files = [os.path.join(gt_dir, f) for f in os.listdir(gt_dir) if os.path.isfile(os.path.join(gt_dir, f))]print(raw_files)
# 执行分析
for raw_file in raw_files:print(f"Analyzing {raw_file}...")# results = analyze_gt_blur(raw_file, height, width, bl)# print(f"Global sharpness variance: {results['global_var']:.2f}\n")results = enhanced_analysis(raw_file,height=height,width=width,bl=bl,output_dir='./enhanced_results')全局清晰度评估
- 拉普拉斯方差指标（Global Laplacian Variance: 97.50）
-
- 质量评级：★★★☆☆（中等偏上）（典型值域：10-200）
- 数值对比参考：
-
- 优质图像：>100
- 可接受范围：50-100
- 需要优化：<50
二、模糊热图特征分析
- 空间分布特征：
| 区域坐标 | 模糊值 | 对应物体 | 成因推测 |
| (620, 880) | 0.82 | 桌面反光区域 | 镜面反射过曝 |
| (1120, 480) | 0.76 | 移动的椅子 | 拍摄时物体位移 |
| (280, 1040) | 0.68 | 墙面装饰画 | 镜头畸变导致 |
