OSCR

Wavelet-enhanced autoencoder based image denoising with CNN fusion.

Code ↔ Paper

6 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 6 matches
  1. [1] § Materials and methods › Proposed method ↔ waveletfusion2303.m, lines 131–188 · score 0.75 · imageInputLayer, regressionLayer, convolution2dLayer, patchSize, Padding, ReLU
  2. [2] § Materials and methods › Proposed method ↔ waveletfusion2303.m, lines 90–129 · score 0.70 · bior3.3, wavelet families, hybrid wavelet, Coif2, Db4, Sym4
  3. [3] § Materials and methods › Proposed method ↔ sotamethods2303.m, lines 87–135 · score 0.66 · imageInputLayer, regressionLayer, convolution2dLayer, Padding, map, ReLU
  4. [4] § Materials and methods › Wavelet ↔ waveletfusion2303.m, lines 90–129 · score 0.63 · bior3.3, wavelet families, Coif2, Db4, Sym4, Haar
  5. [5] § Results and discussion ↔ sotamethods2303.m, lines 1–6 · score 0.55 · SSEQ metric, FFDNet, DnCNN, NIQE, training, Denoising
  6. [6] § Materials and methods › Proposed method ↔ waveletfusion2303.m, lines 131–188 · score 0.51 · Fusion CNN, ReLU layers, regression, AE, patch

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The authors' code

MATLAB · 402 lines · 17 KB · CC-BY-4.0 · 4 matches

  1. clc; clear; close all;
  3. % =========================================================
  4. % FOLDER SELECTION
  5. % =========================================================
  6. imgFolder = uigetdir(pwd, 'Select image folder');
  7. if imgFolder == 0
  8. error('No folder selected. Terminating.');
  9. end
  11. % =========================================================
  12. % COLLECT IMAGE FILES
  13. % =========================================================
  14. imgExt = {'*.jpg','*.png','*.bmp','*.tif'};
  15. fileList = [];
  16. for e = 1:length(imgExt)
  17. fileList = [fileList; dir(fullfile(imgFolder, imgExt{e}))];
  18. end
  20. numImages = length(fileList);
  21. fprintf('Total images found in folder: %d\n', numImages);
  23. if numImages == 0
  24. error('No supported images found in folder!');
  25. end
  27. if numImages < 10
  28. error('Not enough images! At least 10 images required.');
  29. end
  31. % =========================================================
  32. % IMAGE-LEVEL SPLIT: 80% / 10% / 10%
  33. % =========================================================
  34. rng(42);
  35. idx = randperm(numImages);
  36. nTrain = round(0.80 * numImages);
  37. nVal = round(0.10 * numImages);
  39. trainFiles = fileList(idx(1 : nTrain));
  40. valFiles = fileList(idx(nTrain+1 : nTrain+nVal));
  41. testFiles = fileList(idx(nTrain+nVal+1 : end));
  43. fprintf('Train: %d | Validation: %d | Test: %d images\n', ...
  44. length(trainFiles), length(valFiles), length(testFiles));
  46. % =========================================================
  47. % PATCH EXTRACTION — TRAIN AND VAL ONLY (64x64, stride 32)
  48. % =========================================================
  49. patchSize = 64;
  50. patchStride = 32;
  52. fprintf('\nExtracting Train and Val patches...\n');
  53. [YTrain, XTrain] = extractPatches(trainFiles, imgFolder, patchSize, patchStride);
  54. [YVal, XVal ] = extractPatches(valFiles, imgFolder, patchSize, patchStride);
  55. fprintf('Train patches: %d | Val patches: %d\n', size(XTrain,4), size(XVal,4));
  57. % =========================================================
  58. % TEST — LOAD FULL IMAGES (NO PATCHING)
  59. % =========================================================
  60. fprintf('\nLoading test images at full resolution...\n');
  61. nTestImg = length(testFiles); % <-- önce tanımla
  62. testImages = cell(1, nTestImg);
  63. testNoisy = cell(1, nTestImg);
  65. for i = 1:nTestImg
  66. raw = imread(fullfile(imgFolder, testFiles(i).name));
  67. raw = im2double(raw);
  68. if size(raw,3) == 3
  69. raw = rgb2gray(raw);
  70. end
  71. testImages{i} = raw;
  72. testNoisy{i} = imnoise(raw, 'gaussian', 0, 0.01);
  73. end
  74. fprintf('Number of test images: %d\n', nTestImg);
  76. % =========================================================
  77. % AUTOENCODER ARCHITECTURE (Table 4 compliant)
  78. % =========================================================
  79. layersAE = [
  80. imageInputLayer([patchSize patchSize 1],'Name','input','Normalization','none')
  82. % --- Encoder ---
  83. convolution2dLayer(3, 64,'Padding','same','Name','enc_conv1')
  84. reluLayer('Name','enc_relu1')
  85. convolution2dLayer(3, 32,'Padding','same','Name','enc_conv2')
  86. reluLayer('Name','enc_relu2')
  87. convolution2dLayer(3, 16,'Padding','same','Name','enc_conv3')
  88. reluLayer('Name','enc_relu3')
  90. % --- Decoder ---
  91. convolution2dLayer(3, 32,'Padding','same','Name','dec_conv1')
  92. reluLayer('Name','dec_relu1')
  93. convolution2dLayer(3, 64,'Padding','same','Name','dec_conv2')
  94. reluLayer('Name','dec_relu2')
  96. % --- Output layer (Linear activation) ---
  97. convolution2dLayer(3, 1,'Padding','same','Name','output_conv')
  99. regressionLayer('Name','output')
  100. ];
  102. optionsAE = trainingOptions('adam', ...
  103. 'MaxEpochs', 100, ...
  104. 'MiniBatchSize', 32, ...
  105. 'InitialLearnRate', 1e-3, ...
  106. 'ValidationData', {XVal, YVal}, ...
  107. 'ValidationFrequency', 30, ...
  108. 'ValidationPatience', 10, ...
  109. 'Shuffle', 'every-epoch', ...
  110. 'Verbose', false, ...
  111. 'Plots', 'training-progress');
  113. fprintf('\nStarting Autoencoder training...\n');
  114. netAE = trainNetwork(XTrain, YTrain, layersAE, optionsAE);
  115. fprintf('Autoencoder training complete.\n');
  117. % =========================================================
  118. % HYBRID WAVELET DENOISING — TRAIN AND VAL
  119. % =========================================================
  120. waveletFamilies = {'sym4','db4','bior3.3','coif2','haar'};
  121. level = 4;
  123. fprintf('\nComputing Wavelet denoising (train/val)...\n');
  124. YPredAE_Train = predict(netAE, XTrain);
  125. YPredAE_Val = predict(netAE, XVal);
  127. WavTrain = hybridWaveletBatch(XTrain, YTrain, waveletFamilies, level);
  128. WavVal = hybridWaveletBatch(XVal, YVal, waveletFamilies, level);
  129. fprintf('Wavelet denoising complete.\n');
  131. % =========================================================
  132. % FUSION CNN (Table 5 compliant)
  133. % =========================================================
  134. FusTrain = cat(3, YPredAE_Train, WavTrain);
  135. FusVal = cat(3, YPredAE_Val, WavVal);
  137. layersFusion = [
  138. imageInputLayer([patchSize patchSize 2],'Normalization','none','Name','input')
  139. convolution2dLayer(3, 64,'Padding','same','Name','conv1')
  140. reluLayer('Name','relu1')
  141. convolution2dLayer(3, 32,'Padding','same','Name','conv2')
  142. reluLayer('Name','relu2')
  143. convolution2dLayer(3, 16,'Padding','same','Name','conv3')
  144. reluLayer('Name','relu3')
  145. convolution2dLayer(3, 32,'Padding','same','Name','conv4')
  146. reluLayer('Name','relu4')
  147. convolution2dLayer(3, 64,'Padding','same','Name','conv5')
  148. reluLayer('Name','relu5')
  149. convolution2dLayer(3, 1,'Padding','same','Name','conv_out')
  150. regressionLayer('Name','output')
  151. ];
  153. optionsFusion = trainingOptions('adam', ...
  154. 'MaxEpochs', 100, ...
  155. 'MiniBatchSize', 32, ...
  156. 'InitialLearnRate', 1e-3, ...
  157. 'ValidationData', {FusVal, YVal}, ...
  158. 'ValidationFrequency', 30, ...
  159. 'ValidationPatience', 10, ...
  160. 'Shuffle', 'every-epoch', ...
  161. 'Verbose', false, ...
  162. 'Plots', 'training-progress');
  164. fprintf('\nStarting Fusion CNN training...\n');
  165. netFusion = trainNetwork(FusTrain, YTrain, layersFusion, optionsFusion);
  166. fprintf('Fusion CNN training complete.\n');
  168. % =========================================================
  169. % TEST — FULL IMAGE INFERENCE AND METRICS
  170. % =========================================================
  171. fprintf('\nProcessing test images...\n');
  173. psnrNoisy_arr = zeros(1,nTestImg); ssimNoisy_arr = zeros(1,nTestImg);
  174. psnrAE_arr = zeros(1,nTestImg); ssimAE_arr = zeros(1,nTestImg);
  175. psnrWav_arr = zeros(1,nTestImg); ssimWav_arr = zeros(1,nTestImg);
  176. psnrAvg_arr = zeros(1,nTestImg); ssimAvg_arr = zeros(1,nTestImg);
  177. psnrFus_arr = zeros(1,nTestImg); ssimFus_arr = zeros(1,nTestImg);
  178. psnrWiener_arr = zeros(1,nTestImg); ssimWiener_arr = zeros(1,nTestImg);
  180. predAE_full = cell(1, nTestImg);
  181. predWav_full = cell(1, nTestImg);
  182. predAvg_full = cell(1, nTestImg);
  183. predFus_full = cell(1, nTestImg);
  184. predWiener_full = cell(1, nTestImg);
  186. for i = 1:nTestImg
  187. cleanImg = testImages{i};
  188. noisyImg = testNoisy{i};
  190. % Autoencoder only — sliding window
  191. predAE = slidingWindowPredict(netAE, noisyImg, patchSize);
  193. % Wavelet only — full image
  194. predWav = hybridWaveletSingle(noisyImg, cleanImg, waveletFamilies, level);
  196. % Wavelet + AE (average fusion)
  197. predAvg = 0.5 * predAE + 0.5 * predWav;
  199. % Proposed: Wavelet + AE + CNN
  200. predFus = slidingWindowPredictFusion(netFusion, predAE, predWav, patchSize);
  202. % Wiener filter — full image
  203. predWiener = wiener2(noisyImg, [5 5]);
  204. predWiener = max(min(predWiener, 1), 0);
  206. predAE_full{i} = predAE;
  207. predWav_full{i} = predWav;
  208. predAvg_full{i} = predAvg;
  209. predFus_full{i} = predFus;
  210. predWiener_full{i} = predWiener;
  212. psnrNoisy_arr(i) = psnr(noisyImg, cleanImg); ssimNoisy_arr(i) = ssim(noisyImg, cleanImg);
  213. psnrAE_arr(i) = psnr(predAE, cleanImg); ssimAE_arr(i) = ssim(predAE, cleanImg);
  214. psnrWav_arr(i) = psnr(predWav, cleanImg); ssimWav_arr(i) = ssim(predWav, cleanImg);
  215. psnrAvg_arr(i) = psnr(predAvg, cleanImg); ssimAvg_arr(i) = ssim(predAvg, cleanImg);
  216. psnrFus_arr(i) = psnr(predFus, cleanImg); ssimFus_arr(i) = ssim(predFus, cleanImg);
  217. psnrWiener_arr(i) = psnr(predWiener, cleanImg); ssimWiener_arr(i) = ssim(predWiener, cleanImg);
  219. fprintf(' Image %d/%d done\n', i, nTestImg);
  220. end
  222. % =========================================================
  223. % PERFORMANCE REPORT
  224. % =========================================================
  225. fprintf('\n===== AVERAGE TEST SET PERFORMANCE (Full Image) =====\n');
  226. fprintf('%-25s PSNR (dB) SSIM\n','');
  227. fprintf('%-25s %.2f %.4f\n', 'Noisy', mean(psnrNoisy_arr), mean(ssimNoisy_arr));
  228. fprintf('%-25s %.2f %.4f\n', 'Wiener Filter', mean(psnrWiener_arr), mean(ssimWiener_arr));
  229. fprintf('%-25s %.2f %.4f\n', 'Wavelet only', mean(psnrWav_arr), mean(ssimWav_arr));
  230. fprintf('%-25s %.2f %.4f\n', 'Autoencoder only', mean(psnrAE_arr), mean(ssimAE_arr));
  231. fprintf('%-25s %.2f %.4f\n', 'Wav + AE (avg)', mean(psnrAvg_arr), mean(ssimAvg_arr));
  232. fprintf('%-25s %.2f %.4f\n', 'Proposed (Wav+AE+CNN)', mean(psnrFus_arr), mean(ssimFus_arr));
  234. % Improvement rates
  235. fprintf('\n===== IMPROVEMENT RATES =====\n');
  236. fprintf('%-25s PSNR Impr(%%) SSIM Impr(%%)\n','');
  237. methods = {'Wiener Filter','Wavelet only','Autoencoder only','Wav + AE (avg)','Proposed (Wav+AE+CNN)'};
  238. psnrVals = [mean(psnrWiener_arr), mean(psnrWav_arr), mean(psnrAE_arr), mean(psnrAvg_arr), mean(psnrFus_arr)];
  239. ssimVals = [mean(ssimWiener_arr), mean(ssimWav_arr), mean(ssimAE_arr), mean(ssimAvg_arr), mean(ssimFus_arr)];
  240. for m = 1:5
  241. dp = (psnrVals(m) - mean(psnrNoisy_arr)) / mean(psnrNoisy_arr) * 100;
  242. ds = (ssimVals(m) - mean(ssimNoisy_arr)) / mean(ssimNoisy_arr) * 100;
  243. fprintf('%-25s %+.2f%% %+.4f%%\n', methods{m}, dp, ds);
  244. end
  247. % =========================================================
  248. % VISUAL COMPARISON (First 3 test images — main methods)
  249. % =========================================================
  250. numVis = min(3, nTestImg);
  251. for i = 1:numVis
  252. figure('Name',sprintf('Test Image %d',i),'Position',[100 100 1600 350]);
  253. subplot(1,6,1); imshow(testImages{i}); title('Original');
  254. subplot(1,6,2); imshow(testNoisy{i}); title('Noisy');
  255. subplot(1,6,3); imshow(predWav_full{i}); title('Wavelet only');
  256. subplot(1,6,4); imshow(predAE_full{i}); title('AE only');
  257. subplot(1,6,5); imshow(predAvg_full{i}); title('Wav+AE avg');
  258. subplot(1,6,6); imshow(predFus_full{i}); title('Proposed');
  259. sgtitle(sprintf(['Image %d | PSNR — Noisy:%.2f Wav:%.2f ' ...
  260. 'AE:%.2f Avg:%.2f Proposed:%.2f'], ...
  261. i, psnrNoisy_arr(i), psnrWav_arr(i), ...
  262. psnrAE_arr(i), psnrAvg_arr(i), psnrFus_arr(i)));
  263. end
  265. % =========================================================
  266. % WIENER FILTER — SEPARATE VISUAL (First 3 test images)
  267. % =========================================================
  268. for i = 1:numVis
  269. figure('Name',sprintf('Wiener Filter — Test Image %d',i),'Position',[150 150 900 350]);
  270. subplot(1,3,1); imshow(testImages{i}); title('Original');
  271. subplot(1,3,2); imshow(testNoisy{i}); title('Noisy');
  272. subplot(1,3,3); imshow(predWiener_full{i}); title('Wiener Filter');
  273. sgtitle(sprintf('Wiener | Image %d | PSNR: %.2f SSIM: %.4f', ...
  274. i, psnrWiener_arr(i), ssimWiener_arr(i)));
  275. end
  277. % Save results
  278. save('denoising_results.mat', 'netAE', 'netFusion', ...
  279. 'psnrNoisy_arr', 'ssimNoisy_arr', ...
  280. 'psnrAE_arr', 'ssimAE_arr', ...
  281. 'psnrWav_arr', 'ssimWav_arr', ...
  282. 'psnrAvg_arr', 'ssimAvg_arr', ...
  283. 'psnrFus_arr', 'ssimFus_arr', ...
  284. 'psnrWiener_arr','ssimWiener_arr');
  285. fprintf('\nResults saved to denoising_results.mat\n');
  287. % =========================================================
  288. % HELPER FUNCTIONS
  289. % =========================================================
  290. function [cleanPatches, noisyPatches] = extractPatches(fileList, folder, patchSize, patchStride)
  291. cleanPatches = [];
  292. noisyPatches = [];
  293. for i = 1:length(fileList)
  294. raw = imread(fullfile(folder, fileList(i).name));
  295. raw = im2double(raw);
  296. if size(raw,3) == 3
  297. raw = rgb2gray(raw);
  298. end
  299. noisy = imnoise(raw, 'gaussian', 0, 0.01);
  300. [H, W] = size(raw);
  301. for r = 1 : patchStride : H - patchSize + 1
  302. for c = 1 : patchStride : W - patchSize + 1
  303. cp = raw(r:r+patchSize-1, c:c+patchSize-1);
  304. np = noisy(r:r+patchSize-1, c:c+patchSize-1);
  305. cleanPatches = cat(4, cleanPatches, reshape(cp, [patchSize patchSize 1 1]));
  306. noisyPatches = cat(4, noisyPatches, reshape(np, [patchSize patchSize 1 1]));
  307. end
  308. end
  309. end
  310. end
  312. function hybridOut = hybridWaveletBatch(noisySet, cleanSet, waveletFamilies, level)
  313. numWavelets = length(waveletFamilies);
  314. numSamples = size(noisySet, 4);
  315. hybridOut = zeros(size(noisySet));
  316. for s = 1:numSamples
  317. nImg = noisySet(:,:,:,s);
  318. cImg = cleanSet(:,:,:,s);
  319. recons = zeros([size(nImg,1), size(nImg,2), numWavelets]);
  320. psnrVals = zeros(1, numWavelets);
  321. for w = 1:numWavelets
  322. [C, S] = wavedec2(nImg, level, waveletFamilies{w});
  323. detailCoeffs = detcoef2('d', C, S, 1);
  324. sigma = median(abs(detailCoeffs(:))) / 0.6745;
  325. N = numel(detailCoeffs);
  326. T = sigma * sqrt(2 * log(N));
  327. C_thresh = wthresh(C, 's', T);
  328. rec = waverec2(C_thresh, S, waveletFamilies{w});
  329. recons(:,:,w) = rec;
  330. psnrVals(w) = psnr(rec, cImg);
  331. end
  332. [sorted, si] = sort(psnrVals, 'descend');
  333. w1 = sorted(1) / (sorted(1) + sorted(2));
  334. w2 = sorted(2) / (sorted(1) + sorted(2));
  335. hybridOut(:,:,:,s) = w1 * recons(:,:,si(1)) + w2 * recons(:,:,si(2));
  336. end
  337. end
  339. function hybridOut = hybridWaveletSingle(noisyImg, cleanImg, waveletFamilies, level)
  340. numWavelets = length(waveletFamilies);
  341. recons = zeros([size(noisyImg,1), size(noisyImg,2), numWavelets]);
  342. psnrVals = zeros(1, numWavelets);
  343. for w = 1:numWavelets
  344. [C, S] = wavedec2(noisyImg, level, waveletFamilies{w});
  345. detailCoeffs = detcoef2('d', C, S, 1);
  346. sigma = median(abs(detailCoeffs(:))) / 0.6745;
  347. N = numel(detailCoeffs);
  348. T = sigma * sqrt(2 * log(N));
  349. C_thresh = wthresh(C, 's', T);
  350. rec = waverec2(C_thresh, S, waveletFamilies{w});
  351. recons(:,:,w) = rec;
  352. psnrVals(w) = psnr(rec, cleanImg);
  353. end
  354. [sorted, si] = sort(psnrVals, 'descend');
  355. w1 = sorted(1) / (sorted(1) + sorted(2));
  356. w2 = sorted(2) / (sorted(1) + sorted(2));
  357. hybridOut = w1 * recons(:,:,si(1)) + w2 * recons(:,:,si(2));
  358. hybridOut = max(min(hybridOut, 1), 0);
  359. end
  361. function outImg = slidingWindowPredict(net, inImg, patchSize)
  362. [H, W] = size(inImg);
  363. outImg = zeros(H, W);
  364. cntMap = zeros(H, W);
  365. step = patchSize / 2;
  366. for r = 1 : step : H - patchSize + 1
  367. for c = 1 : step : W - patchSize + 1
  368. patch = inImg(r:r+patchSize-1, c:c+patchSize-1);
  369. patch4D = reshape(single(patch), [patchSize, patchSize, 1, 1]);
  370. pred = double(squeeze(predict(net, patch4D)));
  371. outImg(r:r+patchSize-1, c:c+patchSize-1) = ...
  372. outImg(r:r+patchSize-1, c:c+patchSize-1) + pred;
  373. cntMap(r:r+patchSize-1, c:c+patchSize-1) = ...
  374. cntMap(r:r+patchSize-1, c:c+patchSize-1) + 1;
  375. end
  376. end
  377. cntMap(cntMap == 0) = 1;
  378. outImg = outImg ./ cntMap;
  379. outImg = max(min(outImg, 1), 0);
  380. end
  382. function outImg = slidingWindowPredictFusion(net, aeImg, wavImg, patchSize)
  383. [H, W] = size(aeImg);
  384. outImg = zeros(H, W);
  385. cntMap = zeros(H, W);
  386. step = patchSize / 2;
  387. for r = 1 : step : H - patchSize + 1
  388. for c = 1 : step : W - patchSize + 1
  389. pAE = aeImg(r:r+patchSize-1, c:c+patchSize-1);
  390. pWav = wavImg(r:r+patchSize-1, c:c+patchSize-1);
  391. patch4D = reshape(single(cat(3, pAE, pWav)), [patchSize, patchSize, 2, 1]);
  392. pred = double(squeeze(predict(net, patch4D)));
  393. outImg(r:r+patchSize-1, c:c+patchSize-1) = ...
  394. outImg(r:r+patchSize-1, c:c+patchSize-1) + pred;
  395. cntMap(r:r+patchSize-1, c:c+patchSize-1) = ...
  396. cntMap(r:r+patchSize-1, c:c+patchSize-1) + 1;
  397. end
  398. end
  399. cntMap(cntMap == 0) = 1;
  400. outImg = outImg ./ cntMap;
  401. outImg = max(min(outImg, 1), 0);
  402. end

waveletfusion2303.m, under CC-BY-4.0 · at the source

Overview

Authors: Iclal Cetin Tas1
  1. Department of Computer Engineering, Baskent University, Etimesgut, 06790 Ankara, Turkey
Institutions: Başkent University (Türkiye)
Journal: Scientific reports, volume 16, issue 1, article 16503
Dates: received 28 November 2025; accepted 30 March 2026; published online 6 April 2026
Type: Research article · Language: English
License: CC BY
Identifiers: DOI 10.1038/s41598-026-47179-1 · PMID 41942558 · PMCID PMC13216609 · OpenAlex W7151020372
Open access: gold, a free copy (OpenAlex)
Status: code verified
Categories: methods / tools (subfield)
Methods: Machine learning
Keywords: Medical images, Wavelet, Autoencoder, CNN fusion, Noise reduction, Computational biology and bioinformatics, Engineering, Mathematics and computing
Topic: Image and Signal Denoising Methods (Computer Vision and Pattern Recognition, Computer Science), according to OpenAlex
Citations: not cited yet (Europe PMC); 44 references in the paper

Abstract

In biomedical imaging, noise is a fundamental problem negatively impacting diagnostic quality, and effective noise reduction methods are critical for preserving structural details. This study proposes a hybrid noise reduction framework integrated with a CNN-based fusion network, combining a convolutional autoencoder, a specially designed hybrid wavelet filtering approach, and multi-level adaptive wavelet transformation. The autoencoder component learns clean image matches from noisy inputs, while the wavelet-based arm ensures the preservation of high-frequency structural and textural information through multi-level decomposition. The CNN-based fusion module adaptively combines these complementary representations to produce the final enhanced image. Experimental results demonstrate that the proposed method provides consistent superiority across different noise types and datasets. In the fetal ultrasound dataset, the highest performance was achieved at all levels under Gaussian and speckle noise; for speckle noise low-level PSNR values of 48.12 dB and SSIM of 0.99 were obtained. For brain tumor MR dataset, the method was effective not only under Gaussian and speckle noise but also under Rician noise; the PSNR value was obtained as 22.90 dB at the low level, 27.11 dB at the medium level, and 22.17 dB at the high level, with corresponding SSIM values of 0.90, 0.59, and 0.47. Furthermore, in non-referenced evaluations performed on the wrist trauma clinical X-ray dataset, mean NIQE = 3.12 and SSEQ = 0.26 values were obtained. These results confirm that the proposed approach offers balanced and generalizable denoising performance without excessive smoothing or the creation of artificial artifacts on real clinical images. These findings demonstrate that combining wavelet-based detail preservation with deep learning-based reconstruction delivers robust and reliable noise reduction performance across diverse biomedical imaging modalities and noise types.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repository

Its files are read in the Code ↔ Paper reader above, with 6 matches between paragraphs and lines of code.

Zenodo 19081947

License: CC-BY-4.0
State: the link answers, verified on 29 September 2026
Evidence: files inventoried
Languages: MATLAB (3)
Size: 3 files, 3 scripts
Software Heritage: not checked
Found in: “Data availability”
Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation
Tools: Deep Learning Toolbox (2 files), Statistics and Machine Learning Toolbox (2 files), Image Processing Toolbox (1 file), Wavelet Toolbox (1 file)
Availability: 1 check, the latest on 29 September 2026: the link answers (HTTP 200)
  • 29 September 2026: the link answers (HTTP 200)
3 files
At the source: doi.org/10.5281/zenodo.19081947

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

  • 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
  • 3 scripts, each with its path and the digest of its content;
  • 6 matches between paragraphs of the paper and lines of the code (method lexical-v1);
  • neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

Datasets cited

Data availability

The datasets analyzed in this research is publicly accessible third-party resources hosted on Figshare and Kaggle. It can be retrieved from the following link: https://figshare.com/articles/dataset/ brain_tumor_dataset/1512427 (DOI:10. 6084/m9.figshare.1512427.v5).This dataset is distributed under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) license, allowing use, sharing, and reproduction in any format, provided that proper credit is given to the original source. The fetal ultrasound dataset can be accessed from the following link: https://www.kaggle.com/datasets/ankit8467/ fetal-head-ultrasound-dataset-for-image-segment. Third dataset isThe GRAZPEDWRI-DX dataset .It is a publicly available dataset.The dataset is freely accessible under the Creative Commons Attribution 4.0 (CC BY 4.0) licensehttps://www.kaggle.com/datasets/jasonroggy/grazpedwri-dx.The code is avaliable at Zenodo: https://doi.org/10.5281/zenodo.19081947

Reproduced under the paper's license (CC BY), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 29 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 1 author, 8 keywords, 35 references.

Cite

This paper

Cetin Tas, I. (2026). Wavelet-enhanced autoencoder based image denoising with CNN fusion. Scientific reports, 16(1), 16503. https://doi.org/10.1038/s41598-026-47179-1

BibTeX

@article{cetintas2026wavelet,
author = {Cetin Tas, Iclal},
title = {{Wavelet-enhanced autoencoder based image denoising with CNN fusion}},
journal = {Scientific reports},
year = {2026},
month = apr,
volume = {16},
number = {1},
pages = {16503},
publisher = {Nature Publishing Group},
issn = {2045-2322},
doi = {10.1038/s41598-026-47179-1},
url = {https://doi.org/10.1038/s41598-026-47179-1},
pmid = {41942558},
pmcid = {PMC13216609}
}

RIS

TY - JOUR
AU - Cetin Tas, Iclal
TI - Wavelet-enhanced autoencoder based image denoising with CNN fusion
T2 - Scientific reports
J2 - Sci Rep
PY - 2026
DA - 2026/04/06
VL - 16
IS - 1
SP - 16503
SN - 2045-2322
PB - Nature Publishing Group
DO - 10.1038/s41598-026-47179-1
UR - https://doi.org/10.1038/s41598-026-47179-1
LA - en
ER -

CSL-JSON

{
"id": "10.1038/s41598-026-47179-1",
"type": "article-journal",
"title": "Wavelet-enhanced autoencoder based image denoising with CNN fusion",
"container-title": "Scientific reports",
"author": [
{
"family": "Cetin Tas",
"given": "Iclal"
}
],
"container-title-short": "Sci Rep",
"volume": "16",
"issue": "1",
"page": "16503",
"DOI": "10.1038/s41598-026-47179-1",
"PMID": "41942558",
"PMCID": "PMC13216609",
"ISSN": "2045-2322",
"publisher": "Nature Publishing Group",
"URL": "https://doi.org/10.1038/s41598-026-47179-1",
"language": "en",
"issued": {
"date-parts": [
[
2026,
4,
6
]
]
}
}

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.1038/s41467-026-71458-0 [code]
Early differential impact of MeCP2 mutations on functional networks in Rett syndrome patient-derived human cortical organoids.
Journal: Nature communications
In common: Deep Learning Toolbox, Wavelet Toolbox, Image Processing Toolbox, 1 other tool
[2] doi:10.1523/eneuro.0032-26.2026 [code]
Deep Learning Discriminates Seizures from Normal Brain Oscillations in the Electroencephalogram of a Rat Model of Post-traumatic Epilepsy.
Journal: eNeuro
In common: Deep Learning Toolbox, Wavelet Toolbox, Image Processing Toolbox
[3] doi:10.3389/fpsyt.2026.1737357 [code]
A computational pipeline for a neurotransmitter-centric analysis of the effects of psychiatric medication on EEG spectral power.
Journal: Frontiers in psychiatry
In common: Deep Learning Toolbox, Wavelet Toolbox, Statistics and Machine Learning Toolbox
[4] doi:10.3390/diagnostics16152407
BG-YOLO11s: Boundary-Guided YOLO11 with Bézier Contour Augmentation for Brain Tumour Segmentation in T1-CE MRI.
Journal: Diagnostics (Basel, Switzerland)
In common: figshare 1512427, methods / tools, 1 reference
[5] doi:10.1186/s40708-026-00303-3 [code]
BrainFusionNet: a deep learning and XAI model to understand local, global, and sequential features of MRI images for improved brain tumour detection.
Journal: Brain informatics
In common: figshare 1512427, methods / tools, 1 reference
[6] doi:10.1186/s40708-026-00325-x
CAE-BrainNet: a statistically validated class-adaptive attention ensemble model for explainable brain tumor classification from MRI.
Journal: Brain informatics
In common: figshare 1512427, 1 reference
[7] doi:10.1038/s41598-026-52615-3
MANet: a multimodal attention convolutional neural network for brain tumor classification.
Journal: Scientific reports
In common: figshare 1512427, 1 reference
[8] doi:10.1371/journal.pcbi.1014038 [code]
Fast and accessible morphology-free functional fluorescence imaging analysis.
Journal: PLoS computational biology
In common: Wavelet Toolbox, Image Processing Toolbox, Statistics and Machine Learning Toolbox, methods / tools
[9] doi:10.1038/s41467-026-76581-6 [code]
Thalamocortical bursts encode reward contingencies and drive associative learning.
Journal: Nature communications
In common: Deep Learning Toolbox, Image Processing Toolbox, Statistics and Machine Learning Toolbox
[10] doi:10.7554/elife.107661 [code]
In vivo mapping of striatal neurodegeneration in Huntington's disease with Soma and Neurite Density Imaging.
Journal: eLife
In common: Deep Learning Toolbox, Image Processing Toolbox, Statistics and Machine Learning Toolbox

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.