基于形态学的权重自适应图像去噪的MATLAB实现

classdef MorphologicalAdaptiveDenoising% 基于形态学的权重自适应图像去噪类% 支持灰度图像和彩色图像的去噪propertiesstructuring_elementsalphabetaiterationsendmethodsfunction obj = MorphologicalAdaptiveDenoising(varargin)% 构造函数p = inputParser;addParameter(p, 'alpha', 0.7, @(x) x>0 && x<=1);addParameter(p, 'beta', 0.3, @(x) x>0 && x<=1);addParameter(p, 'iterations', 2, @(x) x>0);parse(p, varargin{:});obj.alpha = p.Results.alpha;obj.beta = p.Results.beta;obj.iterations = p.Results.iterations;% 初始化结构元素obj.structuring_elements.small = strel('square', 3);obj.structuring_elements.medium = strel('square', 5);obj.structuring_elements.large = strel('square', 7);endfunction denoised_image = denoise(obj, image, varargin)% 主去噪函数p = inputParser;addParameter(p, 'method', 'adaptive', @ischar);addParameter(p, 'noise_level', 'medium', @ischar);parse(p, varargin{:});if ndims(image) == 3% 彩色图像处理 - 分别处理每个通道denoised_image = zeros(size(image));for channel = 1:3switch p.Results.methodcase 'adaptive'denoised_image(:,:,channel) = ...obj.weighted_morphological_denoising(...image(:,:,channel));case 'advanced'denoised_image(:,:,channel) = ...obj.advanced_morphological_denoising(...image(:,:,channel), p.Results.noise_level);endendelse% 灰度图像处理switch p.Results.methodcase 'adaptive'denoised_image = obj.weighted_morphological_denoising(image);case 'advanced'denoised_image = obj.advanced_morphological_denoising(...image, p.Results.noise_level);endenddenoised_image = uint8(denoised_image);endfunction local_variance = calculate_local_variance(obj, image, window_size)% 计算局部方差if nargin < 3window_size = 5;endif ~isfloat(image)image = im2double(image);end% 使用积分图像加速计算padded_image = padarray(image, [window_size, window_size]/2, 'replicate');squared_image = padded_image .^ 2;% 计算均值和平方均值mean_filter = ones(window_size) / (window_size^2);local_mean = imfilter(padded_image, mean_filter, 'same');local_mean_sq = imfilter(squared_image, mean_filter, 'same');% 计算方差: Var = E[X^2] - (E[X])^2local_variance = local_mean_sq - local_mean .^ 2;% 去除填充pad = floor(window_size/2);local_variance = local_variance(pad+1:end-pad, pad+1:end-pad);endfunction gradient_mag = calculate_gradient_magnitude(obj, image)% 计算梯度幅值if ~isfloat(image)image = im2double(image);end% Sobel算子sobel_x = fspecial('sobel');sobel_y = sobel_x';grad_x = imfilter(image, sobel_x, 'replicate');grad_y = imfilter(image, sobel_y, 'replicate');gradient_mag = sqrt(grad_x.^2 + grad_y.^2);% 归一化if max(gradient_mag(:)) > 0gradient_mag = gradient_mag / max(gradient_mag(:));endendfunction weights = adaptive_weight_calculation(obj, image)% 计算自适应权重if ~isfloat(image)image = im2double(image);end% 计算局部特征local_var = obj.calculate_local_variance(image);gradient_mag = obj.calculate_gradient_magnitude(image);% 归一化local_var_norm = (local_var - min(local_var(:))) / ...(max(local_var(:)) - min(local_var(:)) + eps);gradient_norm = (gradient_mag - min(gradient_mag(:))) / ...(max(gradient_mag(:)) - min(gradient_mag(:)) + eps);% 计算权重：细节区域权重小，平坦区域权重大weights = 1.0 ./ (1.0 + obj.alpha * local_var_norm + obj.beta * gradient_norm);endfunction results = morphological_operations(obj, image, se_type)% 执行多种形态学操作switch se_typecase 'small'se = obj.structuring_elements.small;case 'medium'se = obj.structuring_elements.medium;case 'large'se = obj.structuring_elements.large;endresults.opening = imopen(image, se);results.closing = imclose(image, se);results.gradient = imdilate(image, se) - imerode(image, se);results.top_hat = imtophat(image, se);results.black_hat = imbothat(image, se);endfunction denoised_image = weighted_morphological_denoising(obj, image)% 权重自适应形态学去噪if ~isfloat(image)image = im2double(image);enddenoised = image;for iter = 1:obj.iterations% 计算自适应权重weights = obj.adaptive_weight_calculation(denoised);% 转换到uint8进行形态学操作denoised_uint8 = im2uint8(denoised);% 多尺度形态学操作results_small = obj.morphological_operations(denoised_uint8, 'small');results_medium = obj.morphological_operations(denoised_uint8, 'medium');results_large = obj.morphological_operations(denoised_uint8, 'large');% 转换回doublefield_names = fieldnames(results_small);for i = 1:length(field_names)field = field_names{i};results_small.(field) = im2double(results_small.(field));results_medium.(field) = im2double(results_medium.(field));results_large.(field) = im2double(results_large.(field));end% 权重融合weighted_result = zeros(size(denoised));% 根据权重分类区域mask_high_weight = weights > 0.7;      % 平坦区域mask_medium_weight = weights >= 0.3 & weights <= 0.7; % 过渡区域mask_low_weight = weights < 0.3;       % 细节区域% 平坦区域：使用大结构元素weighted_result(mask_high_weight) = ...0.6 * results_large.opening(mask_high_weight) + ...0.4 * results_large.closing(mask_high_weight);% 过渡区域：使用中等结构元素weighted_result(mask_medium_weight) = ...0.4 * results_medium.opening(mask_medium_weight) + ...0.4 * results_medium.closing(mask_medium_weight) + ...0.2 * results_small.gradient(mask_medium_weight);% 细节区域：使用小结构元素，保留原始细节weighted_result(mask_low_weight) = ...0.5 * results_small.opening(mask_low_weight) + ...0.3 * results_small.closing(mask_low_weight) + ...0.2 * denoised(mask_low_weight);denoised = weighted_result;enddenoised_image = denoised;endfunction denoised_image = advanced_morphological_denoising(obj, image, noise_level)% 增强版形态学去噪if nargin < 3noise_level = 'medium';endif ~isfloat(image)image = im2double(image);end% 根据噪声水平调整参数switch noise_levelcase 'low'iter = 1;alpha_temp = 0.5;beta_temp = 0.2;case 'medium'iter = 2;alpha_temp = 0.7;beta_temp = 0.3;case 'high'iter = 3;alpha_temp = 0.9;beta_temp = 0.4;end% 临时修改参数original_alpha = obj.alpha;original_beta = obj.beta;obj.alpha = alpha_temp;obj.beta = beta_temp;% 形态学重建reconstructed = obj.morphological_reconstruction_denoising(image, iter);% 权重计算和最终融合weights = obj.adaptive_weight_calculation(reconstructed);denoised_image = obj.final_fusion(image, reconstructed, weights);% 恢复原始参数obj.alpha = original_alpha;obj.beta = original_beta;endfunction reconstructed = morphological_reconstruction_denoising(obj, image, iterations)% 形态学重建去噪marker = image;for i = 1:iterations% 交替开闭运算opening = imopen(marker, obj.structuring_elements.small);closing = imclose(marker, obj.structuring_elements.small);% 重建操作marker = imclose(opening, obj.structuring_elements.medium);marker = imopen(closing, obj.structuring_elements.medium);endreconstructed = marker;endfunction fused = final_fusion(obj, original, denoised, weights)% 最终结果融合if ~isfloat(original)original = im2double(original);endif ~isfloat(denoised)denoised = im2double(denoised);end% 计算边缘保护系数gradient_original = obj.calculate_gradient_magnitude(original);edge_preserve = exp(-gradient_original * 5);% 融合fused = edge_preserve .* original + (1 - edge_preserve) .* denoised;endendmethods (Static)function noisy_image = add_noise(original_image, noise_type, varargin)% 添加噪声switch noise_typecase 'gaussian'p = inputParser;addOptional(p, 'sigma', 25, @isnumeric);parse(p, varargin{:});sigma = p.Results.sigma / 255;noise = sigma * randn(size(original_image));noisy_image = im2double(original_image) + noise;noisy_image = im2uint8(noisy_image);case 'salt_pepper'p = inputParser;addOptional(p, 'density', 0.05, @isnumeric);parse(p, varargin{:});noisy_image = imnoise(original_image, 'salt & pepper', p.Results.density);case 'poisson'noisy_image = imnoise(original_image, 'poisson');endendfunction psnr_value = calculate_psnr(original, denoised)% 计算PSNRif ~isfloat(original)original = im2double(original);endif ~isfloat(denoised)denoised = im2double(denoised);endmse_value = mean((original(:) - denoised(:)) .^ 2);if mse_value == 0psnr_value = inf;elsepsnr_value = 20 * log10(1 / sqrt(mse_value));endendfunction ssim_value = calculate_ssim(original, denoised)% 计算SSIMssim_value = ssim(denoised, original);endend
end% 演示函数
function demonstrate_morphological_denoising()% 演示形态学去噪效果% 创建测试图像original = create_test_image();% 添加噪声noisy = MorphologicalAdaptiveDenoising.add_noise(original, 'gaussian', 30);% 创建去噪器denoiser = MorphologicalAdaptiveDenoising('alpha', 0.7, 'beta', 0.3, 'iterations', 2);% 执行去噪tic;denoised_basic = denoiser.denoise(noisy, 'method', 'adaptive');time_basic = toc;tic;denoised_advanced = denoiser.denoise(noisy, 'method', 'advanced', 'noise_level', 'medium');time_advanced = toc;% 计算指标psnr_noisy = MorphologicalAdaptiveDenoising.calculate_psnr(original, noisy);psnr_basic = MorphologicalAdaptiveDenoising.calculate_psnr(original, denoised_basic);psnr_advanced = MorphologicalAdaptiveDenoising.calculate_psnr(original, denoised_advanced);ssim_noisy = MorphologicalAdaptiveDenoising.calculate_ssim(original, noisy);ssim_basic = MorphologicalAdaptiveDenoising.calculate_ssim(original, denoised_basic);ssim_advanced = MorphologicalAdaptiveDenoising.calculate_ssim(original, denoised_advanced);% 显示结果figure('Position', [100, 100, 1200, 800]);subplot(2, 3, 1);imshow(original);title('原始图像');subplot(2, 3, 2);imshow(noisy);title(['加噪图像 - PSNR: ', num2str(psnr_noisy, '%.2f'), ' dB, SSIM: ', num2str(ssim_noisy, '%.3f')]);subplot(2, 3, 3);imshow(denoised_basic);title(['基本去噪 - PSNR: ', num2str(psnr_basic, '%.2f'), ' dB, SSIM: ', num2str(ssim_basic, '%.3f'), ...' (', num2str(time_basic, '%.2f'), 's)']);subplot(2, 3, 4);imshow(denoised_advanced);title(['增强去噪 - PSNR: ', num2str(psnr_advanced, '%.2f'), ' dB, SSIM: ', num2str(ssim_advanced, '%.3f'), ...' (', num2str(time_advanced, '%.2f'), 's)']);% 显示权重图weights = denoiser.adaptive_weight_calculation(noisy);subplot(2, 3, 5);imshow(weights, []);title('自适应权重图');colorbar;% 显示局部方差local_var = denoiser.calculate_local_variance(noisy);subplot(2, 3, 6);imshow(local_var, []);title('局部方差图');colorbar;fprintf('去噪结果对比:\n');fprintf('噪声图像 - PSNR: %.2f dB, SSIM: %.3f\n', psnr_noisy, ssim_noisy);fprintf('基本去噪 - PSNR: %.2f dB, SSIM: %.3f, 时间: %.2fs\n', psnr_basic, ssim_basic, time_basic);fprintf('增强去噪 - PSNR: %.2f dB, SSIM: %.3f, 时间: %.2fs\n', psnr_advanced, ssim_advanced, time_advanced);
endfunction test_image = create_test_image()% 创建测试图像% 方法1: 使用内置图像% test_image = imread('cameraman.tif');% 方法2: 创建合成测试图像test_image = 128 * ones(256, 256, 'uint8');% 添加几何形状test_image(50:100, 50:100) = 200;  % 亮方块test_image(150:200, 150:200) = 60; % 暗方块% 添加纹理[x, y] = meshgrid(1:256, 1:256);texture = uint8(50 * sin(x/10) .* cos(y/10) + 128);test_image = test_image + texture;% 添加边缘test_image(100:110, :) = 0;  % 水平线test_image(:, 100:110) = 0;  % 垂直线test_image = uint8(test_image);
end% 批量处理函数
function batch_denoising_test()% 批量测试不同噪声水平的去噪效果original = create_test_image();noise_levels = [15, 25, 50];methods = {'adaptive', 'advanced'};results = cell(length(noise_levels), length(methods));figure('Position', [50, 50, 1500, 600]);for i = 1:length(noise_levels)noise_sigma = noise_levels(i);noisy = MorphologicalAdaptiveDenoising.add_noise(original, 'gaussian', noise_sigma);subplot(3, 4, (i-1)*4 + 1);imshow(original);title('原始图像');subplot(3, 4, (i-1)*4 + 2);imshow(noisy);title(['噪声图像 (σ=', num2str(noise_sigma), ')']);for j = 1:length(methods)denoiser = MorphologicalAdaptiveDenoising();denoised = denoiser.denoise(noisy, 'method', methods{j});subplot(3, 4, (i-1)*4 + 2 + j);imshow(denoised);psnr_val = MorphologicalAdaptiveDenoising.calculate_psnr(original, denoised);ssim_val = MorphologicalAdaptiveDenoising.calculate_ssim(original, denoised);title([methods{j}, ' - PSNR: ', num2str(psnr_val, '%.2f'), ...', SSIM: ', num2str(ssim_val, '%.3f')]);results{i, j} = denoised;endend
end% 使用示例
% demonstrate_morphological_denoising();
% batch_denoising_test();

这个MATLAB实现包含了以下主要特性：

功能

1. 自适应权重计算

   • 基于局部方差和梯度幅值
   • 自动识别平坦区域和细节区域

2. 多尺度形态学操作

   • 小、中、大三种结构元素
   • 开运算、闭运算、梯度等操作

3. 权重融合策略

   • 平坦区域：使用大结构元素
   • 细节区域：使用小结构元素
   • 过渡区域：中等结构元素

4. 增强版算法

   • 形态学重建
   • 边缘保护融合

使用

% 基本使用
denoiser = MorphologicalAdaptiveDenoising();
denoised_image = denoiser.denoise(noisy_image);% 高级选项
denoised_image = denoiser.denoise(noisy_image, 'method', 'advanced', 'noise_level', 'high');% 彩色图像处理
denoised_color = denoiser.denoise(color_image);

参考代码 基于形态学的权重自适应图像去噪 www.youwenfan.com/contentcni/63976.html

优势

1. 自适应性强：根据图像局部特征自动调整处理策略
2. 细节保护：在去噪的同时有效保护图像边缘和纹理
3. 多尺度处理：结合不同尺度的形态学操作
4. 参数可调：支持灵活的参数调整以适应不同噪声水平