A Deeper Understanding of Luminance and Chrominance for Computer Vision

Computer vision texts often describe edge detection as an operation on grayscale images. The grayscale versions of images are often found as the Y component of Y C
_B
C
_R
, or the V component of HSV color spaces. Students are taught that Y channel of Y C
_B
C
_R
is the same as luminance, which they understand as whiteness or brightness. Two fundamental problems result from this misunderstanding. First, students develop the misconception that an image with zero value of Y in Y C
_B
C
_R
has no colorfulness. Second students develop a cognitive bias, an implicit assumption, that edge detection is always performed in the luminance plane of an image. These misconceptions can persist throughout careers. When feeding a neural network with an image, no single channel is sufficient to detect edges and textures. Here we present a simple demonstration, and a simple counter example, to expose these misconceptions with the goal of developing a deeper understanding of luminance and chrominance. Exercises given to students to circumvent these assumptions are also described.