Why is RGB Euclidean distance a poor measure of perceptual color difference?

WHAT THIS TESTS: This question probes whether you understand that RGB is a device-oriented encoding rather than a model of human vision. Interviewers want to see that you know perceptual uniformity matters when algorithms must align with human judgment, and that you can name and justify a color space designed for that purpose.

A GOOD ANSWER COVERS: First, explain that RGB is linear with respect to light intensity but not with respect to human perception, so a Euclidean distance of ten units in a dark blue region can look like a huge shift while the same distance in a bright green region looks nearly invisible. Second, introduce CIELAB explicitly, noting that L encodes perceptual lightness and a and b encode opponent red-green and blue-yellow axes. Third, state that CIELAB was designed to be perceptually uniform, meaning a given numerical delta corresponds to a roughly similar perceived change across the space, even though it is not perfectly uniform. Fourth, connect this to segmentation by noting that clustering algorithms like k-means or threshold-based edge detection rely on distance metrics, and using LAB makes those distances correlate with actual visual boundaries rather than arbitrary hardware values.

COMMON WRONG ANSWERS: A major red flag is saying RGB distance is fine because it works for simple computer graphics tasks. Another is describing HSV or HSL as perceptually uniform; they are cylindrical transforms of RGB and are not designed for uniform perceptual deltas. A third is claiming CIELAB is just a nonlinear RGB without explaining the opponent color model or the goal of perceptual uniformity. Confusing LAB with the older Hunter Lab or with CMYK also signals shallow knowledge.

LIKELY FOLLOW-UPS: An interviewer might ask how you would compute a perceptual distance metric like Delta E in practice, or why LAB is still not perfectly uniform and what newer spaces like CIECAM02 or IPT address. They might also ask how gamma correction in sRGB relates to the problem, or whether you would convert to LAB for a neural network loss function.

ONE CONCRETE EXAMPLE: Imagine segmenting a medical image where subtle color shifts indicate tissue boundaries. In RGB, a small numerical change near black might be exaggerated while a large change near white is compressed, causing the algorithm to miss a critical boundary. By converting to CIELAB first, a fixed clustering threshold of five units means roughly the same visual difference whether the region is light or dark, so the segmentation mask aligns with what a clinician actually sees.

Source: Wikipedia: CIELAB color space