### 可见光与红外图像融合 可见光与红外图像融合是一种常见的多模态数据处理技术，在遥感、计算机视觉等领域有着广泛应用。为了实现这一目标，可以采用多种算法和技术。 #### 使用OpenCV库进行简单的加权平均法融合 一种简单而有效的方式是对两幅图像执行像素级别的线性组合： ```python import cv2 import numpy as np def weighted_average_fusion(visible_img_path, ir_img_path): visible_image = cv2.imread(visible_img_path, 0) ir_image = cv2.imread(ir_img_path, 0) fused_image = (visible_image.astype(float) + ir_image.astype(float)) / 2 return fused_image.clip(0, 255).astype(np.uint8) fused_result = weighted_average_fusion('path_to_visible.jpg', 'path_to_ir.png') cv2.imshow('Fused Image', fused_result) cv2.waitKey(0) cv2.destroyAllWindows() ``` 这种方法虽然直观易懂，但在实际应用中可能无法很好地保留原始特征信息[^1]。 #### 基于频域变换的方法——离散余弦变换(DCT) 更高级别的方法涉及频率域操作，比如DCT变换能够更好地保持细节并减少伪影现象： ```python from skimage import io, transform, img_as_float import pywt def dct_based_fusion(vis_img, inf_img): vis_float = img_as_float(io.imread(vis_img)) inf_float = img_as_float(io.imread(inf_img)) # Resize images to same size if necessary min_shape = sorted([(np.sum(i.size), i.size) for i in [vis_float, inf_float]])[0][1] vis_resized = transform.resize(vis_float, min_shape) inf_resized = transform.resize(inf_float, min_shape) coeffs_vis = pywt.dwt2(vis_resized, 'haar') coeffs_inf = pywt.dwt2(inf_resized, 'haar') LL_v, (LH_v, HL_v, HH_v) = coeffs_vis LL_i, (LH_i, HL_i, HH_i) = coeffs_inf # Fusion rule can be modified according to specific needs. LL_fused = (LL_v + LL_i)/2 LH_fused = (abs(LH_v)+abs(LH_i))/2 HL_fused = (abs(HL_v)+abs(HL_i))/2 HH_fused = (abs(HH_v)+abs(HH_i))/2 coeffs_fused = LL_fused, (LH_fused, HL_fused, HH_fused) reconstructed_image = pywt.idwt2(coeffs_fused,'haar') return reconstructed_image fusion_output = dct_based_fusion('visible_image_file', 'infrared_image_file') io.imshow(fusion_output) io.show() ``` 上述代码展示了基于Haar小波的二维离散余弦变换来完成图像间的融合过程[^2]。