Convert 24 bit 2D array to 8 bit 3D (RGB) array

Hello all,

I am reading in data from a camera sensor, and it is stored as a 32-bit 2D array. Here, the first 8 bits correspond to Red, the next 8 bits to Green, and so on. I want to split this 2D array (of size (a,b)) to a 3D array (a,b,3), where each element in the array is 8-bit long. I am currently doing this as follows:

[r c]= size(img);

IMG = zeros(r,c,3);

 for i = 1:r

     for j = 1:c

         temp = dec2bin(img(i,j),24);

         IMG(i,j,1) = bin2dec(fliplr(temp(1:8)));
         IMG(i,j,2) = bin2dec(fliplr(temp(9:16)));
         IMG(i,j,3) = bin2dec(fliplr(temp(17:24)));
    end
 end
 figure, imshow(IMG(:,:,1))
 figure, imshow(IMG(:,:,2))
 figure, imshow(IMG(:,:,3))

This takes a very long time to compute (~ 20 minutes) for the large images that I am currently using.

Is there a more efficient way of doing this?

Thanks in advance!

[Jan]

What is the 4th channel?

[Oshin]

The 4th channel doesn't have any relevant information. It's all zeros.

Answer 1

[Walter Roberson]

IMGB = reshape( typecast(img, 'uint8'), [4, size(img)]);

Then row 1 will correspond to one of the bands, row 2 to another, etc..

You might need to do a bit of work to figure out which row is which.

[Jan]

If speed matters (does it ever?), James' typecast function is recommended: http://www.mathworks.com/matlabcentral/fileexchange/17476-typecast-and-typecastx-c-mex-functions

[Geoff]

Good point. I use MatLab for prototyping speed, not execution speed.

Answer 2

[Geoff]

Try using bitwise operators like you would in C. These handily operate on a matrix, so you can split out your components like so:

[r c]= size(img);
IMG = zeros(r,c,3);

IMG(:,:,1) = uint8(bitand(img, 255))
IMG(:,:,2) = uint8(bitand(bitshift(img,-8), 255));
IMG(:,:,3) = uint8(bitand(bitshift(img,-16), 255));

[Geoff]

Hmmm I edited this to add the uint8 wrapper, but it might not behave quite right. MatLab's type conversion (say from signed to unsigned) doesn't work the way _I_ would expect.

[Jan]

I assume that a division is faster than the bit-shifting.

[Geoff]

Well, I come from a C background so I just assumed that those functions share some kind of parallel. =) If division in MatLab is faster, that doesn't say much about the bitshift function!