noise removal from the part of the medical image where noise is actually present, not the whole image

i am doing one algorithm where i need to remove noise from an image , but i cant apply any kind of filter or any noise removal algorithm directly to the whole image since i am working in medical image and it might blur or might take any tissue or connected organ as a noise and remove it and will change my output. i would like to know the best possible way to remove noise from only those parts of the image where noise actually exists. thank you..

[Kian Azami]

Can you upload an image as an example?

[syeda musfia]

this is an MR image of brain

Answer 1

[Bjorn Gustavsson]

If you know what regions of the images you have noise that needs filtering why not just filter those regions? Something like this (one-region example):

regN = [i11,i12,i21,i22]; % corners of noisy region
ImgN = Im_full((regN(1)-1):(regN(2)+1),(regN(3)-1):(regN(4)+1)); Will break down for regions that extends to image edges, fix to suit needs
ImgN = wiener2(ImgN,[3,3]); % Or whatever filtering suits your images
Im_full(regN(1):regN(2),regN(3):regN(4)) = ImgN(2:end-1,2:end-1);

To me it sounds strange that you need to reduce noise in what seems to be "medically uninteresting" parts of images, while the parts of the images with organs and tissue should be left unfiltered with noise intact. Why not simply tell the doctors that the noisy parts of the images are left noisy to make them aware of the general noise characteristics of the images, that should be information that should help image interpretation.

HTH

[syeda musfia]

i dont know which part of my image contains the noise, i need to find those areas then apply filter in those areas only. and i want to remove any kind of noise because the noisy parts can be confused with any disease in the image.

[Bjorn Gustavsson]

Well, then it sounds like you have a very difficult problem. Not the actual filtering itself but how to determine what is noise and what is images of disease-affected tissue. If you cant explain to yourself what separates the noise from the signal there's no way we can do it for you.

I could go on and guess a lot but that would be wild shots in the dark.

For example I could guess that the image noise has no spatial correlation, i.e. noise-signal is independent random signals on a pixel-to-pixel scale, then different simple low-pass filters with very narrow filter-kernels would be the first thing to try, but the disease-affected tissue might also have spatial variations at the highest spatial frequencies, so that might also be a bad idea. Have a look at wiener2, medfilt2, look for bilateral filters and non-linear diffusion filters on the file exchange, determine what the characteristics of your noise are. Unless you can answer those questions you wouldn't know whether any of our suggestions are good or terrible...

HTH

[learningmat]

hi..in my case, i can detect the noisy part and i can remove, but i dont know how to index it and replace those pixels with noise free image. Initially am splitting the image into blocks, each block am applying some threshold to detect the noisy part, if it is affected, removing it by applying weiner filter to that particular block, but i dont know how to replace it with noise free image. Thanks in advance

[Image Analyst]

Nirmala, in my answer below, I told how you use a mask to replace the identified bad/noisy pixels with the repaired, noise-reduced pixels. Perhaps you overlooked it.

[Walter Roberson]

All parts of MRI images have noise. Guaranteed.

[learningmat]

thank you for the reply sir..but i want to reduce the complexity so i dont want to apply the filtering to whole image, i want to apply to only the noisy part. Actually i can apply it for noisy part alone but how to indexing it in an array and replacing it with noisy part

[Image Analyst]

Again, SEE MY ANSWER where I already told you how.

Or CLICK HERE if you don't want to scroll down with your mouse wheel.

Answer 2

[Kian Azami]

You can do something as follow which I learned from one of the MathWorks STAFF:

close all
% Read the
I = imread('input.jpg');
% Remove noise
se1 = strel('disk',1);
I1 = imclose(I, se1);
se2 = strel('disk',2)
I2 = imopen(I1, se2);
% Original Image
figure
imshow(I)
% Image after noise
figure
imshow(I2)

Basically, imopen(I,se1) will remove the pixels that are left alone.

Then, imclose(I1,se2) will do the opposite. It will unite the pixels with the same color.

se1 and se2 are the structures that tell imopen and imclose how to remove or unite the pixels.

By the command 'strel' you define the zone by which you can remove or unite the pixels. For example, strel('disk',2) is a disk like zone with a dimension of 2. So, when you apply this structure in your image by 'imopen' it will divide the image into disk like shapes with the dimension of 2 and if it see discontinuity with the picture colors it will remove the pixels in that zone which is possibly a noise. You can play with the dimensions and see the change in your image.

You can define many different kinds of strel and if you go to help you can see them: strel Documentation

Here I put the link of the similar question that I asked before: Noise Removal

Answer 3

[Image Analyst]

I agree with Bjorn - the radiologist will be able to mentally filter out the noisy areas.

If you still want to do it for some reason, then you'll have to find a mask for identifying the noisy areas (for example use stdfilt() but exclude edge regions), then remove noise for the entire image, and replace the noise pixels with noise free pixels.

mask = some algorithm to identify noisy regions.
noiseFreeImage = ReduceNoise(grayImage); % Remove noise over the entire image.
noiseReducedImage = grayImage; % Initialize.
% Replace pixels in noise region with noise-free pixels.
noiseReducedImage (mask) = noiseFreeImage(mask);