issue in converting matrix to rgb image

I was trying to replace pixel values by their corresponding 3x3 non overlapping block. since the output matrix was of type double I converted it to uint8 type.

But when I displayed the image in the figure window it was gray instead of coloured image . I tried cat(3, gray, gray, gray) but image is still colourless.

What did I do wrong?

 img= imread('C:\Users\prach\MATLAB programs\Lenna_(test_image).png');
 new = blockproc(img, [10 10], fun);
 new1= uint8(new);
 imshow(new1)
 rgbnew= cat(3, new1, new1, new1);
 imshow(rgbnew)

[PRACHI Sood]

I forgot to add the definitiion of fun. Here it is

fun = @(block_struct) meadian(block_struct.data);

[Subhadeep Koley]

@PRACHI Your original "Lenna_(test_image).png" is gray-scale or color image?

[Walter Roberson]

Remember that median() works along the first dimension by default, so median() of a 10 x 10 block is a 1 x 10 block. Unless, that is, the block is rgb, in which case median() of a 10 x 10 x 3 block is a 1 x 10 x 3 block. You probably want median(block_struct.data, [1 2])

Notice also that you are not doing a sliding window, so your output would be about 10 time smaller than the input.

[PRACHI Sood]

@Subhadeep original image is a color image

[PRACHI Sood]

@Walter yes, reducing spatial resolution was my aim. I will try it with median(block_struct.data, [1 2]) also. Thankyou for the information.

edit: I tried the alternative but it throws the error:

"error using median (line 94)

Dimension argument must be a positive integer scalar within indexing range."

[Walter Roberson]

sounds like you are using a slightly older version of MATLAB that does not permit multiple dimensions. In that case as long as the image grayscale, median(block_struct.data(:))

But if you are dealing with rgb then it gets a bit messier,

median(reshape(block_struct.data,[],1,3),1)

[PRACHI Sood]

@Walter Roberson

The median(block_struct.data(:)) worked perfectly. Thankyou very much.

Answer 1

[Image Analyst]

You need to use imsplit() to get the individual color channels, and then process them and combine them with cat(3).

[PRACHI Sood]

I tried this solution. But it gives the error :

"Unable to perform assignment because the size of the left side is 171-by-171 and the size of the right side is

128-by-128."

I tried to solve this but failed. Following is my code.

new = blockproc(img, [3 3], fun);
new1= uint8(new);      
rgbim(:,:,1)= new1;
rgbim(:,:,2)= new1;
rgbim(:,:,3)= new1;
subplot(2,4,n-1), imshow(rgbim);

[Image Analyst]

Attach the m-file and image. All you gave is a small partial snippet that can't run. How are you defining fun()???

See my attached blockproc() demos.

[Image Analyst]

The blockproc_color_means.m demo gives this image. Is it like you want?

[PRACHI Sood]

Apologies for the incomplete information. Attaching the .m file and the image.

Yes, the output images shown in your last comment as output of blockproc_color_means.m is my desired output. I will go through the list of demos you have provided. Thankyou very much.