How to apply same 2D logical mask to each 3D layer effctively?

RGB = imread('onion.png');
mask = RGB(:,:,1)>100;
%%How can I do something like below
RGB(mask,1:3) = 0;       % I want to apply same 2D 'mask' for all the 3 layers

I got this intuition from this post "logical operation on a column within a matrix" .

Answer 1

[Image Analyst]

Try this:

% Mask the image using bsxfun() function to multiply the mask by each channel individually.
maskedRgbImage = bsxfun(@times, rgbImage, cast(mask, 'like', rgbImage));

[JAI PRAKASH]

Hey IA Thanx

In continuation, can u tell me how can I perform below operation. Because thats my final target.

Note: Mask can be created with any of R G B layer, because black areas are completely black, i.e. '0' for each R G B's. In this way efficiency can be improved.

[Image Analyst]

Try this (untested);

mask = Im2 ~= 0;  % Any non-pure-black pixels)
OverlappedIm = im1; % Initialize to Im1
% Now replace with pixels from Im2:
OverlappedIm(mask) = Im2(mask);

[JAI PRAKASH]

Ya it is working accurately.

But my concern is why to make 'mask' a 3D matrix if each layer will be having same data.

Can't this code structure be make more efficient!!

[Image Analyst]

That is the most efficient way. It uses linear indexing to assign ONLY those pixels that need to have their color reassigned, unlike the multiplication method with bsxfun() where every pixel has to undergo a multiplication. It should be faster to reassign a small fraction of the pixels than to multiply all the pixels.

[JAI PRAKASH]

ya, above code is straight forward and fast.

But the below code I find 10% more faster

[m, n, p] = size(Im1);
maskIdx = find(Im2(:,:,1) > 0);
maskIdx = [maskIdx; m*n+maskIdx; 2*m*n+maskIdx];
Im1(maskIdx) = Im2(maskIdx);  % Im1 will become the OverlappedIm

frankly say, I still hope for further efficiency..

[Image Analyst]

It might be slightly faster but it's not general. It's depending on there being at least some red in every pixel of Im2 that you want to use as a mask. So it won't work for a general color where the shape in Im2 could be anything but would work in your specific computer graphics case.

[Walter Roberson]

Using the multiply version can be faster sometimes because the multiply version on sufficiently large matrices can be sent to LAPACK for multithreaded execution.