Compare YCbCr and RGB. Why chroma subsampling for compression?

What's really being asked

This question probes whether you understand why video and image pipelines transform pixel data into a different color space before compression. The interviewer wants to see that you know RGB is not optimal for perceptual coding, that YCbCr provides a decorrelated representation, and that you can connect engineering decisions like chroma subsampling to biological constraints of the human visual system rather than treating them as arbitrary tricks.

The full answer

A strong response should hit four points in order. First, describe RGB as an additive color model where all three channels carry both luminance and chrominance information, making them highly correlated and thus inefficient for compression. Second, explain that YCbCr is derived from RGB primaries but separates the signal into one luma channel (Y) and two chroma channels (Cb and Cr), which decorrelates the data and lets compressors treat brightness and color independently. Third, state the perceptual principle: the human retina contains roughly 120 million rods versus 6 to 7 million cones, giving us far higher spatial resolution for luminance than for color; therefore, our brains reconstruct fine detail primarily from brightness edges, not color edges. Fourth, connect this directly to subsampling by noting that standards like 4:2:2 or 4:2:0 store chroma at half or quarter resolution, reducing raw bandwidth by roughly 33% to 50% while preserving perceived sharpness because the dropped color samples are below our perceptual threshold.

The mistakes people make

One red flag is saying chroma is subsampled because it is simply less important than luma without mentioning spatial acuity; this signals a memorized fact without understanding. Another mistake is claiming YCbCr is used because it reduces file size automatically, ignoring that the transform itself is lossless and the savings come from downstream quantization and subsampling. Some candidates also confuse YCbCr with YPBPR, so be ready to note that YCbCr is the digital quantization of the same underlying concept.

What usually comes next

An interviewer might push on the math by asking you to compute the byte difference between a 1920x1080 RGB24 frame and a 4:2:0 YCbCr frame, which is roughly 6.2 MB versus 3.1 MB. They could also ask why subsampling is less acceptable for computer screen text or chroma keying, where sharp color edges matter. Another angle is asking how modern codecs like HEVC or AV1 handle chroma beyond simple subsampling, such as cross-component linear model prediction.

A concrete example

Consider a 4K video stream at 3840 by 2160. In uncompressed RGB8, each frame requires 3840 times 2160 times 3 bytes, about 24.9 MB. Converting to YCbCr does not change the uncompressed size, but applying 4:2:0 subsampling drops the two chroma planes to 1920 by 1080 each. The resulting frame is then about 12.4 MB, a 50% reduction before any transform coding or quantization even occurs, and viewers typically cannot distinguish the subsampled version from full resolution in natural imagery.