Removing noise from images and keep them in RGB format.

Hi

I want to remove noise from a RGB image. There are two type of problem i am facing. 1- Couldn't find a right tool for that in Matlab. 2- The available tool like Median filter works on the RGB images.

What is the solution for this to remove the noise and keep them in the RGB format or at least bring them back to this format.

Kindly guide me.

Regards

Answer 1

[Image Analyst]

You can extract the color channels separately then do a median filter on each channel with medfilt2(), then recombine with cat() into a new RGB image.

[Ghulam]

Hi,

Thanks for your reply. I have applied the median filter onto my image my there was no meaningful results. I think that this type of noise couldn't be removed using the Median Filter.

I am sharing my code here for your reference so if there is any error in implementing it you can point out.

RGB = imread('1.jpg');
ch1 = RGB(:,:,1);
ch2 = RGB(:,:,2);
ch3 = RGB(:,:,3);
ch1 = medfilt2(ch1);
ch2 = medfilt2(ch2);
ch3 = medfilt2(ch3);
final = cat(3,ch1,ch2,ch3);
imshow(final)
end

I am also attaching the actual and processed pictures and mentioning the area i want to remove.

Original Image

Processed Image after the application of the above code

Thanks for your reply. Have a nice day.

[Image Analyst]

Use a larger window size and you should see some difference. It's the second input argument to medfilt2().

[Ghulam]

Thanks a lot for your reply

I have implemented the median filter with window fortunately its removing the noise but it is also removing the important information from the image.

Below is the code and image which can give you a better idea of my implementation.

RGB = imread('1.jpg');
ch1 = RGB(:,:,1);
ch2 = RGB(:,:,2);
ch3 = RGB(:,:,3);
ch1 = medfilt2(ch1,[80 80]);
ch2 = medfilt2(ch2,[80 80]);
ch3 = medfilt2(ch3,[80 80]);
final = cat(3,ch1,ch2,ch3);
figure; subplot(1,2,1); imshow(RGB)
subplot(1,2,2); imshow(final)
imwrite(final,'result.jpg');

Here is the result of the filter.

Here you can easily view that the blood vessels are removed completely and the actual information is not present now.

Could you please suggest me how to retain the valuable information at one hand and remove the noise on the other hand.

Please let me know about it.

Regards

[Image Analyst]

See ophthalmology articles here that tell you how to do it:

http://www.visionbib.com/bibliography/contents.html

Check out the section on medical imaging and look for the retina/eye/ophthalmology section.

[Ghulam]

Thanks a lot for your continuous reply.

I have found this link which has solved the same problem but has used Mathematica for coding.

I have tried using this solution in matlab as shown here.

rgb = imread('7.jpg');
gray = rgb2gray(rgb);
bin = imbinarize(gray,0.5);
figure; subplot(1,5,1); imshow(rgb);
subplot(1,5,2); imshow(gray);
subplot(1,5,3);imshow(bin)
SE = strel('disk',20);
J = imdilate(bin,SE);
subplot(1,5,4); imshow(J);
K = roifill(gray,J);
subplot(1,5,5); imshow(K);
figure
imshow(K)

But couldn't find how to implement inpaint in Matlab.

I have used roifill() instead of inpaint as suggested by you in other posts but it removes the complete area. Also it only works in gray images which is another issue.

I am sharing here the processed image too.

Many thanks to you.

[Image Analyst]

You can try regionfill().

[Ghulam]

Thanks once again for your reply.

I also tried it using regionfill(), but unfortunately its giving the same results.

I am attaching the image of the result.

Regards

[Image Analyst]

Ghulam, it's not a simple thing. Again, see ophthalmology articles here that tell you how to do it:

http://www.visionbib.com/bibliography/contents.html

Check out the section on medical imaging and look for the retina/eye/ophthalmology section. You're not going to be able to do it with some simple 30 line program.

If you want an overview of denoising methods, see https://users.soe.ucsc.edu/~milanfar/publications/journal/ModernTour.pdf State of the art algorithms include BM3D, KSVD, non-local means, K-LLD, UINTA, and so on. Simple filters like linear filtering or even non-linear filtering like median filtering just can't perform as well as the more sophisticated ones. So if the simple median is not meeting your needs, use a more sophisticated one.

[Ghulam]

Really thankful for your support.