Improve Speed to Convert an RGB Image Stored as a Linear Array to a Matrix

I have an RGB image stored as a vector 921,000 bytes long (640x480x3 uint8) in a sequence as follows:

[B(1,1) G(1,1) R(1,1) B(1,2) G(1,2) R(1,2) B(1,3) G(1,3) R(1,3) ... B(480,640) G(480,640) R(480,640)]

I used a FOR loop to convert it to a matrix as shown below. I would like to speed-up this conversion. Any suggestions?

    r = zeros(480,640,'uint8');
    g = r;
    b = r;
    a = my_image_obj.Bytes.uint8;
    for y = 1:480
        f = 1920*(y-1);
        for x = 1:640
            s = 3*x-2+f;
            b(y,x) = a(s);
            g(y,x) = a(s+1);
            r(y,x) = a(s+2);
        end
    end
    my_image(:,:,1) = r;
    my_image(:,:,2) = g;
    my_image(:,:,3) = b;

Answer 1

[Tim Zaman]

use reshape: http://www.mathworks.com/help/techdoc/ref/reshape.html

to reshape to [B(11) G(11) B(11) ; ... ] so that it is now [3x307000] the first column is now B, second G, third B.

Now, use reshape again to reshape to a square size. (almost) never use for loops!

[Sonoma Rich]

I tried the following and it works. Any further suggestions to improve it?

a = my_image_obj.Bytes.uint8;

b = reshape(a,3,length(a)/3);

my_image(:,:,1) = reshape(b(3,:),640,480)';

my_image(:,:,2) = reshape(b(2,:),640,480)';

my_image(:,:,3) = reshape(b(1,:),640,480)';

[Jan]

Improvement: Pre-allocation, e.g. by starting with my_image(:, :, 3).

Answer 2

[Sean de Wolski]

A one-liner

my_image = reshape(fliplr(reshape(a,3,numel(a)/3).'),[640 480 3])

[Sonoma Rich]

I tried this but the image is rotated 90 degrees. The image is 640 horizontal by 480 vertical. This line produces a 480 horizontal by 630 vertical image.

[Sean de Wolski]

my_image = permute(my_image,[2 1 3]);

[Sean de Wolski]

or

doc imrotate

doc rot90