### 图像融合方法概述 图像融合是一种将多张图片的信息结合起来的技术，通常用于增强视觉效果或提取特定特征。金字塔融合和小波变换融合是两种常见的图像融合技术。 #### 1. **金字塔融合** 金字塔融合的核心思想在于构建高斯金字塔和拉普拉斯金字塔来分解图像的不同频率分量，并在不同层次上进行融合后再重构图像[^1]。具体过程包括以下几个方面： - 使用 `cv2.pyrDown` 和 `cv2.pyrUp` 函数生成高斯金字塔。 - 计算拉普拉斯金字塔作为高频细节的保留部分。 - 将两张图像在同一层上的对应区域按需组合（如苹果左侧与橘子右侧），并逐步重建最终图像。 #### 2. **小波变换融合** 小波变换能够更精细地捕捉图像的空间频率特性。它通过对原图应用离散小波变换 (DWT)，分离低频近似系数以及水平、垂直和对角方向的高频细节[^3]。随后可以根据需求选取合适的策略（比如取平均值或者最大绝对值）完成像素级融合操作，最后逆向执行 IDWT 得到合成结果。 以下是基于 Python 的实现代码： ```python import cv2 import numpy as np from pywt import dwt2, idwt2 def gaussian_pyramid(image, levels=6): lower = image.copy() pyramid_images = [lower] for i in range(levels - 1): lower = cv2.pyrDown(lower) pyramid_images.append(lower) return pyramid_images def laplacian_pyramid(gaussian_pyr): laplacian_pyr = [] for i in range(len(gaussian_pyr)-1, 0, -1): size = (gaussian_pyr[i - 1].shape[1], gaussian_pyr[i - 1].shape[0]) expanded = cv2.pyrUp(gaussian_pyr[i], dstsize=size) L = cv2.subtract(gaussian_pyr[i - 1], expanded) laplacian_pyr.append(L) laplacian_pyr.append(gaussian_pyr[-1]) return laplacian_pyr[::-1] def blend_laplace(laplacian_apple, laplacian_orange, mask_pyr): blended_pyr = [] for apple_level, orange_level, mask_level in zip(laplacian_apple, laplacian_orange, mask_pyr): blended = apple_level * mask_level + orange_level * (1.0 - mask_level) blended_pyr.append(blended.astype(np.float32)) return blended_pyr def reconstruct_from_laplacian(blend_levels): reconstructed_image = blend_levels[-1] for level in reversed(blend_levels[:-1]): size = (level.shape[1], level.shape[0]) reconstructed_image = cv2.pyrUp(reconstructed_image, dstsize=size) reconstructed_image = cv2.add(level, reconstructed_image) return reconstructed_image.clip(0, 255).astype(np.uint8) # 加载图像 apple = cv2.imread('apple.jpg').astype(float) / 255. orange = cv2.imread('orange.jpg').astype(float) / 255. mask = np.zeros(apple.shape[:2]).astype(float) mask[:, :int(mask.shape[1]/2)] = 1 levels = 6 # 创建高斯金字塔 gp_apple = gaussian_pyramid(apple, levels) gp_orange = gaussian_pyramid(orange, levels) gp_mask = gaussian_pyramid(mask, levels) # 创建拉普拉斯金字塔 lp_apple = laplacian_pyramid(gp_apple) lp_orange = laplacian_pyramid(gp_orange) # 融合拉普拉斯金字塔 blended_lp = blend_laplace(lp_apple, lp_orange, gp_mask) # 重构成图像 result = reconstruct_from_laplacian(blended_lp) # 显示结果 cv2.imshow("Blended Image", result) cv2.waitKey(0) cv2.destroyAllWindows() # 小波变换融合函数定义 def wavelet_fusion(img1, img2, mode='db1', level=1): coeffs_img1 = dwt2(img1, mode, level=level) cA1, (cH1, cV1, cD1) = coeffs_img1 coeffs_img2 = dwt2(img2, mode, level=level) cA2, (cH2, cV2, cD2) = coeffs_img2 # 取均值或其他方式融合 cA = (cA1 + cA2)/2 cH = (cH1 + cH2)/2 cV = (cV1 + cV2)/2 cD = (cD1 + cD2)/2 fused_coeffs = cA, (cH, cV, cD) fused_img = idwt2(fused_coeffs, mode) return fused_img fused_wavelet_result = wavelet_fusion(apple, orange) cv2.imshow("Wavelet Fused Image", fused_wavelet_result) cv2.waitKey(0) cv2.destroyAllWindows() ``` 以上代码实现了两种不同的图像融合方案——一种采用传统的金字塔融合算法，另一种则运用了小波变换的方式来进行图像数据的综合处理[^2]。