How to get a dithered color image

I know that

X = dither(RGB, map);

returns an indexed image approximation to the given RGB image. I have two questions:

(1) Is there a second step that converts the indexed image to an RGB image using the provided colormap?

(2) How to design a colormap that matches the original colors of the image?

Note: I want a dithered color image not an indexed image.

Any hint/help is appreciated!

Answer 1

[Image Analyst]

I've never used dither() so I don't know what it does. But to answer your questions:

(1) Yes. It's called ind2rgb().

rgbImage = ind2rgb(indexedImage, customColorMap);

(2) Call rgb2ind():

[indexedImage, customColorMap] = rgb2ind(grayImage, numColors);

[elham sa]

Thank you very much image analyst!

The rgb2ind() function was really helpful, now I know how to get the colormap that matches the image.

However, the ind2rgb() returns an RGB continuous image, not a dithered one (I mean the color image with the dots!)

I tried dithering in each separate color channel and it did the trick.

Thanks!

[DGM]

If you don't already have a map appropriate for a given RGB image, then I don't see that there's any reason to be using dither() at all.

% the inputs
nlevels = 16; % maximum number of unique colors
inpict = imread('peppers.png');

% rgb2ind() defaults are dithered already
% if you need to use rgb2ind() to get the optimized map
% then there's no point in ever calling dither()
[idx map] = rgb2ind(inpict,nlevels); 
imshow(idx,map,'border','tight')
xlim([242 388]); ylim([155 274]) % zoom in

% ind2rgb() cannot change the presence or absence of dithering
% so if the output isn't dithered, then either the input was never dithered
% or it's simply being displayed incorrectly somehow
outpict = ind2rgb(idx,map);
imshow(outpict,'border','tight')
xlim([242 388]); ylim([155 274]) % zoom in

% there are at most (nlevels) colors
allcolors = unique(reshape(outpict,[],3),'rows');
numcolors = size(allcolors,1)

numcolors = 16

There's no way that dithering channels independently returns an appropriate result, or a result compatible with any optimized color table. I have no idea what was attempted, but I can guess.

% create an effectively 3-bit color image by
% naively calling dither() on each channel
junk3b = zeros(size(inpict),'uint8');
for c = 1:3
    junk3b(:,:,c) = im2uint8(dither(inpict(:,:,c)));
end

imshow(junk3b,'border','tight')
xlim([242 388]); ylim([155 274]) % zoom in

% there are at most 8 colors
allcolors = unique(reshape(junk3b,[],3),'rows');
numcolors = size(allcolors,1)

numcolors = 8

Or less likely:

% do graylevel dithering on each channel
% dither() doesn't really have a good way to do this
junkgl = zeros(size(inpict),'uint8');
CT = gray(nlevels); % dummy map
for c = 1:3
    thischannel = repmat(inpict(:,:,c),[1 1 3]); % expand
    junkgl(:,:,c) = dither(thischannel,CT);
end
junkgl = im2uint8(double(junkgl)/(nlevels-1)); % rescale
imshow(junkgl,'border','tight')
xlim([242 388]); ylim([155 274]) % zoom in

% ... and there are far more unique colors than specified (at most nlevels^3)
allcolors = unique(reshape(junkgl,[],3),'rows');
numcolors = size(allcolors,1)

numcolors = 1079