Aliasing in Chroma Resampling - Image Processing

Hi there,

I recently heard that while downsampling a 4:4:4 YCbCr data to 4:2:0/4:2:2/4:1:1 formats, if pixels are just dropped - it gives rise to aliasing.

I am not able to figure out how it happens ? Is there an example/mathematical explanation that would suffice the claim ?

THnak you

Hi;
May be the reason is something like this;
Consider an artificial high freq image like one pixel width color/black dots
Color/Black/Color/Black
In 444 it is pixel format would be
Y0Cb0Cr0| Y1Cb1Cr1|Y2Cb2Cr2|Y3Cb3Cr3
If you just drop Cb and Cr component for 422 the pixel format would be like below
(in standard realignment, macro pixels)
Cb0Y0Cr0Y1|Cb2Y2Cr2Y3
(see 4CC website YUV pixel formats)
So when you want to reconstruct RGB image, black pixels (Y1 and Y3) have some color inheritance from the colored neighbor pixels (when you are performing upsampling back to 444 to display it).

Here then arises a new question, how should be performed a good 444 to 422 conversion (or 422 to 444 upsampling scheme). I don’t know the exact methods applied on video systems. Actually this example is on an artificial (very high frequency) picture. So in real world examples it can be ignored, may be…

Hope an expert clarifies.
Good luck.

Hi emresel,

Thank you. Now I am able to visualize high freq component(neighbouring color-black alternate dots) getting converted to a low frequency component- either black only or color only on reconstruction when pixels are just dropped which is the principle of aliasing.

THnak you and regards.