How to apply Median FIlter To a color image with out using 'rgb2gray' ? Program please.

How to add noise and eliminate it using filters (in this case salt and pepper and Median filters respectively) by using without the conversion to gray scale.?

Answer 1

[Guillaume]

Just loop over the colour channels:

processedimage =  zeros(size(yourimage), 'like', yourimage));
for channel = 1:3
   processedimage(:, :, channel) = somefunction(yourimage(:, :, channel));
end

Answer 2

[Image Analyst]

See my attached demos for applying salt and pepper noise to color and gray scale images and then removing it. It should do exactly what you want in a fancy demo. Adapt it as needed.

[RMT]

Hello Image Analyst,

I know it's an old question but i'm taking a course in computer vision and i have a concern that is related to this question.

regarding your method of isolating the R, G, and B channles then process them by the 2d filter then reconstruct the image (i didn't mention that you actually replaced the noisey pixels by those filtered ones because it's not my concern here). My concern is that why you haven't used the 3d Median filter that is available?

Actually I tried your method and in parallel also used the 3d filter then compared the results on the bit level and the two results were different although the same filter settings were used. So can you guid me on this issue?

I mea, for color images, which method is more accurate? the seperation or using the 3d filter? And this 3d filter is it meant for color images or for general 3d images?

Thanks

[Image Analyst]

Well I guess you could - if you were willing to accept more of a color change. The different color channels can have drastically different values. Imagine you were looking at a tomato with values of (200, 40, 20) with salt and pepper noise. Consider if the 3-D 3x3x3 filter window was processing the green channel so it also brings in values of the red and blue channels into the moving window. So if we ignore values of 0 and 255 and take the median value of what's left, you might be left with a value of 200 for the green instead of 40, depending on how much noise was in each channel. Or it might get a value of 20 if the blue channel was the channel with the least noise. So you're assigning colors from a different color channel to it, which would dramatically change the color if the color is not a neutral shade. Doing it one color channel at a time would avoid that. I suppose the red and blue channels might be ok with the 3-D median IF the window shrunk as it got to the edge of the array, meaning you basically ignore values that are off the edge of the image. (I'm not sure how the median filter deals with edge effects - the common ways are to shrink the window or assume values outside the array are zero). But in that case (shrinking window), it's reverted to a 2-D filter window on one color channel alone. So, bottom line, I think doing it channel-by-channel will give more accurate colors.

[RMT]

Thanks a lot for your response.

Best regards.