当前位置：首页 > news >正文

基于区域的空间域图像融合MATLAB实现

news 2026/5/12 1:08:13

基于区域的空间域图像融合MATLAB实现，结合多尺度区域分割与特征加权策略

一、方法原理

区域分割
采用SLIC超像素分割算法将图像划分为均匀区域，每个区域包含约200-300个像素。
区域特征提取
- 颜色特征：HSV空间均值
- 纹理特征：灰度共生矩阵(GLCM)的对比度、相关性
- 结构特征：区域紧凑度（面积/周长²）
区域匹配
使用改进的SSIM算法计算区域相似度，构建区域关联图。
加权融合规则
根据区域显著性设计自适应权重：

其中λ控制方差权重，β调节SSIM影响因子。

二、MATLAB实现代码

%% 参数设置
img1 = imread('visible.png');  % 可见光图像
img2 = imread('infrared.png'); % 红外图像
block_size = 16;              % 分块尺寸
overlap = 4;                  % 重叠区域
lambda = 0.1;                 % 方差抑制系数
beta = 0.5;                   % SSIM权重%% 图像预处理
img1_gray = rgb2gray(img1);
img2_gray = rgb2gray(img2);
[height, width] = size(img1_gray);%% 区域分割
blocks1 = superpixels(img1_gray, 'Compactness', 10, 'NumSuperpixels', height*width/(block_size^2));
blocks2 = superpixels(img2_gray, 'Compactness', 10, 'NumSuperpixels', height*width/(block_size^2));%% 区域特征提取
features1 = region_features(img1, blocks1, block_size);
features2 = region_features(img2, blocks2, block_size);%% 区域匹配
sim_matrix = compute_similarity(features1, features2);%% 区域融合
fused_img = zeros(height, width);
for i = 1:size(blocks1,1)[y1, x1] = ind2sub(size(blocks1), i);y1 = y1*block_size - overlap;x1 = x1*block_size - overlap;% 获取重叠区域roi1 = imcrop(img1, [x1 y1 block_size block_size]);roi2 = imcrop(img2, [x1 y1 block_size block_size]);% 计算区域权重weight1 = exp(-lambda * var(roi1(:))) + beta * ssim(roi1, roi2);weight2 = 1 - weight1;% 加权融合fused_roi = double(roi1)*weight1 + double(roi2)*weight2;fused_img(y1:y1+block_size-1, x1:x1+block_size-1) = fused_roi;
end%% 后处理
fused_img = medfilt2(fused_img, [3 3]);
imshow(fused_img);%% 辅助函数
function features = region_features(img, blocks, block_size)[rows, cols] = size(img);num_blocks = size(blocks,1);features = struct('color',[],'texture',[],'shape',[]);for i = 1:num_blocks[y, x] = ind2sub(size(blocks), i);y = y*block_size - block_size +1;x = x*block_size - block_size +1;block = imcrop(img, [x y block_size block_size]);features(i).color = mean(block(:));features(i).texture = graycomatrix(block,'NumLevels',16);features(i).shape = regionprops(block, 'Area', 'Perimeter');end
endfunction sim = compute_similarity(f1, f2)num_blocks = size(f1,1);sim = zeros(num_blocks);for i = 1:num_blocksfor j = 1:num_blocks% 颜色相似度color_diff = norm(f1(i).color - f2(j).color);% 纹理相似度glcm1 = f1(i).texture;glcm2 = f2(j).texture;contrast1 = graycoprops(glcm1, 'Contrast');contrast2 = graycoprops(glcm2, 'Contrast');texture_diff = abs(contrast1.Contrast - contrast2.Contrast);% 形状相似度area1 = f1(i).shape.Area;area2 = f2(j).shape.Area;shape_diff = abs(log(area1) - log(area2));% 综合相似度sim(i,j) = exp(-0.5*(color_diff^2 + texture_diff^2 + shape_diff^2));endend
end