1 简介
The removal of mixed noise is a stiff problem since the distribution of the noise cannot be predicted accurately. The most common mixed noise is the combination of Additive White Gaussian Noise (AWGN) and Impulse Noise (IN). Many methods first attempt to remove IN but it might collapse the texture of the image. In this paper, we propose a new learning-based method using convolutional neural network (CNN) for removing mixed gaussian-impulse noise. Since our denoising network can remove various level of mixed noise, neither the preprocessing for removing IN nor noise-level estimation is necessary.
2 部分代码
% Use this code when trainingtraining_image_list = {};for a=1000:2:1200 training_image_list = [training_image_list,['image/train/train_image_gray', num2str(a) ,'.png']]; endpatch_size = 33;train_data = zeros(patch_size, patch_size, 1000000, 'single');train_label = zeros(patch_size, patch_size, 1000000, 'single');num_patches = 0;% Make training datafor image_index = 1 : length(training_image_list) fprintf('Reading %s\n', training_image_list{image_index}); for impulse_loop =1 for impulse_noise_rate = 0:5:45 for gaussian_noise_sigma = 0:10:50 img_original = im2single(imread(training_image_list{image_index})); img_original = padarray(img_original, ceil(size(img_original)/patch_size)*patch_size-size(img_original) ,'symmetric','post'); % AWGN img_noisy = img_original + (gaussian_noise_sigma / 255) * randn(size(img_original)); % RVIN img_noise_position = rand(size(img_original)) < impulse_noise_rate / 100; img_noisy(repmat(img_noise_position,1,1)) = rand(1, sum(img_noise_position(:))); if impulse_noise_rate>0 if gaussian_noise_sigma >0 if rand<0.15 % SPIN img_noise_position = rand(size(img_original)) < randi(30) / 100; img_noisy(repmat(img_noise_position,1,1)) = rand(1, sum(img_noise_position(:)))>0.5; end end end tmp_data = im2col(img_noisy, [patch_size, patch_size], 'distinct'); tmp_data = reshape(tmp_data, patch_size, patch_size, []); tmp_label = im2col(img_original, [patch_size, patch_size], 'distinct'); tmp_label = reshape(tmp_label, patch_size, patch_size, []); train_data(:, :, num_patches + 1 : num_patches + size(tmp_data, 3)) = tmp_data; train_label(:, :, num_patches + 1 : num_patches + size(tmp_label, 3)) = tmp_label; num_patches = num_patches + size(tmp_data, 3); end end endendtrain_data = train_data(:, :, 1 : num_patches);train_label = train_label(:, :, 1 : num_patches);% reshape to MxNx1xCtrain_data = reshape(train_data, patch_size, patch_size, 1, []);train_label = reshape(train_label, patch_size, patch_size, 1, []);% save('mixed_data.mat','train_data','train_label');% clear allfprintf('Complete.\n');
3 仿真结果
4 参考文献
Ryo Abiko, and Masaaki Ikehara. "Blind Denoising of Mixed Gaussian-impulse Noise by Single CNN." ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019.
