### Python OpenCV 多波段图像融合 示例代码 多波段图像融合是一种通过组合来自不同源的多个图像来创建单个高质量图像的技术。下面是一个基于Python和OpenCV库实现多波段图像融合的例子。 #### 准备工作 为了执行此操作，需要准备两张或多张相同尺寸的输入图像作为不同的频带[^2]。确保安装了必要的依赖项： ```bash pip install opencv-python numpy matplotlib ``` #### 加载图像 加载待融合的两幅或多幅图像，并转换为浮点数类型以便后续处理。 ```python import cv2 import numpy as np def load_images(image_paths): images = [] for path in image_paths: img = cv2.imread(path, cv2.IMREAD_GRAYSCALE).astype(np.float32) images.append(img) return images ``` #### 构建高斯金字塔与拉普拉斯金字塔 构建每张图像对应的高斯金字塔以及由其派生出来的拉普拉斯金字塔。 ```python def build_pyramids(images, levels=6): gaussian_pyr = [images.copy()] laplacian_pyr = [] for i in range(levels - 1): next_level = [] for j in range(len(gaussian_pyr[i])): downsampled = cv2.pyrDown(gaussian_pyr[i][j]) upsampled = cv2.pyrUp(downsampled, dstsize=(gaussian_pyr[i][j].shape[1], gaussian_pyr[i][j].shape[0])) laplacian = cv2.subtract(gaussian_pyr[i][j], upsampled) next_level.append(downsampled) laplacian_pyr.append(laplacian) gaussian_pyr.append(next_level) return gaussian_pyr[-1], laplacian_pyr[::-1] ``` #### 融合过程 将各层拉普拉斯金字塔对应位置相加得到最终的结果图。 ```python def fuse_laplace(pyr_list): fused_levels = [] num_bands = len(pyr_list[0]) for level_idx in range(num_bands): band_sum = sum([pyramid[level_idx] for pyramid in pyr_list]) / float(len(pyr_list)) fused_levels.insert(0, band_sum) reconstructed_image = reconstruct_from_laplace(fused_levels) return reconstructed_image def reconstruct_from_laplace(pyramid): result = pyramid[0] for i in range(1, len(pyramid)): size = (result.shape[1]*2, result.shape[0]*2) expanded = cv2.resize(result, dsize=size, interpolation=cv2.INTER_LINEAR) result = cv2.add(expanded, pyramid[i]) return result.clip(min=0., max=255.).astype('uint8') ``` #### 主程序入口 定义主函数用于调用以上各个部分完成整个流程。 ```python if __name__ == "__main__": paths = ["image1.jpg", "image2.jpg"] # 替换成实际路径 imgs = load_images(paths) gaus_base, lapa_lists = zip(*[build_pyramids([img]) for img in imgs]) fusion_result = fuse_laplace(list(lapa_lists)) cv2.imshow("Fusion Result", fusion_result) cv2.waitKey(0) cv2.destroyAllWindows() ``` 这段代码展示了如何利用OpenCV中的`cv2.pyrDown()` 和 `cv2.pyrUp()` 方法分别降低分辨率（下采样）和增加分辨率（上采样），并通过减法运算获得细节信息——即所谓的“拉普拉斯差”。最后再把这些差异叠加起来重建出新的合成图像[^1]。