Downsampling a single RGB Channel

How would you go about downsampling a single RGB Channel (R, G or B) and then re-forming the RGB image? I'm fairly new to image processing. I've looked at a few sources, but I haven't managed to find the answer to this in Matlab yet.

For example, if I split the image into its individual R, G and B channels, downsample the G and B channel with sample rate of 2, how would I recombine the R channel with the downsampled G and B channels? They are all now different matrix dimensions. I hope this isn't a silly question and that my example is clear enough to understand.

Answer 1

[Image Analyst]

Use cat

smallRGB = cat(3, smallRedChannel, smallGreenChannel, smallBlueChannel);

Of course you can use use imresize(rgbImage, 0.5) on the original RGB image if you want, rather than splitting it up into separate color channels.

[stayfrosty]

Sorry that's not it. I only want to downsample 2 of the channels. So, I can't concatenate all the channels together because of the differing dimensions.

[Image Analyst]

That's correct, they can't be combined unless they're all the same size. If you want, you can blow up the small images with pixel replication using the 'nearest' option of imresize(). Then you can recombine them and see two of the channels looking kind of blocky.

[stayfrosty]

Thank you very much for the reply and solution. I just want to elaborate on the problem a bit more. I'm actually trying to reproduce the methodology in the attached screenshot.

http://imgur.com/7oMJ1ZF

So, I've converted RGB to the YUV color space, then downsampled U and V. Performed DCT then quantization. The final steps to retrieve the processed image have been left out of the article and I'm kinda doing a bit of guesswork here.

So, now I'm guessing I would perform IDCT then use Matlab's built-in function of imresize() to replicate the missing pixels.

Do you think that's what the researchers in that article would have done? I've tried e-mailing the people who did that research but no reply at the moment.

[Image Analyst]

Notice that they do not combine the different sized images into another image like you asked about. They probably save them separately in binary in some file, one after the other.

To reconstruct, yes, you'd read in the data, then reverse the process, including IDCT and resizing.