i get black image danoise when the simulation finish what is the problem in the code ?

clc

clear all

close all

image_org=imread('C:\Users\pc\OneDrive\Desktop\memoire\chapitre 3\image for test','JPG');

% Gaussian white noise

X = double(image_org) / 255;

NG = imnoise(X, 'gaussian', 0, 0.01);%NG=gaussian noise

figure(1);subplot(231); imshow(image_org);title ('image org')

subplot(233); imshow(NG) ; title ('Gaussian additive noise')

NG_gray = rgb2gray(NG);

%averaging filter

f_m = fspecial ('average', 3)%f_m=averaging filter

NGmoy = imfilter (NG, f_m,'replicate');

%gaussian filter

f_g = fspecial ('gaussian',3,0.8)%f_g=gaussian filter

NGg = imfilter(NG,f_g,'replicate');

%median filter

NGmed = medfilt2( NG_gray ,[3 3]);

%the show

subplot(234); imshow(NGmoy);title(' avraging Filter ')

subplot(235); imshow(NGg);title(' gaussien Filtre ')

subplot(236); imshow(NGmed);title(' median Filtre')

%PSNR

EQMb=mean2((X-NG).^2 );PSNRb=-10*log10(EQMb);

EQMmoy=mean2( (NG-NGmoy).^2 );PSNRmoy=-10*log10(EQMmoy);

EQMg=mean2( (NG-NGg).^2 );PSNRg=-10*log10(EQMg);

EQMmed=mean2( (NG_gray-NGmed).^2 );PSNRmed=-10*log10(EQMmed);

% Resize the original image to match the input size of the CNN

image_resized = imresize(image_org, [299 450]);

% Add Gaussian noise to the original image

X = double(image_resized) / 255;

NG = imnoise(X, 'gaussian', 0, 0.03); % Add Gaussian noise

% Display the original and noisy images

figure(1);

subplot(1, 2, 1);

imshow(image_resized);

title('Original Image');

subplot(1, 2, 2);

imshow(NG);

title('Noisy Image');

% Prepare data for training (noisy images as input, original images as target)

inputData = NG; % Noisy images

targetData = X; % Original clean images

% Define the CNN architecture for image denoising

layers = [

imageInputLayer([299 450 3]) % Input layer with the size of the input images

convolution2dLayer(3, 64, 'Padding', 'same') % Convolutional layer with 64 filters of size 3x3

reluLayer % ReLU activation layer

convolution2dLayer(3, 64, 'Padding', 'same') % Another convolutional layer

reluLayer % ReLU activation layer

convolution2dLayer(3, 3, 'Padding', 'same') % Output convolutional layer with 3 channels (RGB)

regressionLayer % Regression layer

];

% Define training options

options = trainingOptions('adam', ... % Adam optimizer

'MaxEpochs', 50, ... % Increase the number of epochs

'MiniBatchSize', 16, ... % Decrease the mini-batch size

'InitialLearnRate', 1e-4, ... % Decrease the initial learning rate

'Plots', 'training-progress'); % Plot training progress

% Train the network

net = trainNetwork(inputData, targetData, layers, options);

% Denoise the image using the trained CNN

denoisedImage = predict(net, inputData);

% Display the original noisy image and the denoised image

figure;

subplot(1, 2, 1);

imshow(inputData);

title('Noisy Image');

subplot(1, 2, 2);

imshow(denoisedImage);

title('Denoised Image');

[Josh]

might want to return to this line:

denoisedImage = predict(net, inputData);

the output of predict() will not be an image as your code assumes it to be:

https://www.mathworks.com/help/stats/classificationtree.predict.html

you're probably getting a black image for denoisedImage because it's being handled like a BW or boolean image after predict() outputs something different than what you're expecting

[Josh]

...seems like you're trying to do some stuff manually that's available here as well:

https://www.mathworks.com/help/images/deep-learning.html?s_tid=CRUX_lftnav

Answer 1

[Drishti]

Hi Nassim,

The issue of black denoised image you are encountering can be resolved by setting the ‘Normalization’ parameter as ‘none’ in the ‘imageInputLayer’ function.

Setting normalization as ‘none’ will allow the network to utilize the critical information present in the original data.

Use the following code snippet to set the normalization parameter as 'none':

imageInputLayer([299 450 3], 'Normalization', 'none') 

Furthermore, I have attached the output denoised image below, received after modifying the provided code on sample image:

For better understanding, you can refer to the MATLAB documentation of the imageInputLayer function:

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.imageinputlayer.html

I hope this resolves the issue.