grayscale image

A grayscale image carries brightness but no color: each pixel is a single number on a scale from black to white. Strip the hue out of a photograph and what remains is the grayscale image — a map of how much light each point sent to the camera. Because there is one value per pixel instead of three, it is the leanest meaningful image, a single 2D matrix.

Converting color to grayscale is a luminance computation, not a plain average of the channels, because the human eye is far more sensitive to green than to red or blue. The standard weighted formula (ITU-R BT.601) is Y = 0.299 R + 0.587 G + 0.114 B, with green weighted heaviest so the result matches perceived brightness; modern HD content uses the BT.709 weights (0.2126, 0.7152, 0.0722). For physical correctness this weighting should be applied to linear RGB, though many libraries apply it directly to gamma-encoded values as a fast approximation. The output is one channel where 0 is black and the maximum (255 in 8-bit) is white.

Grayscale matters because a great many vision operators care about structure — edges, gradients, corners, texture — which lives in brightness, not color. Classical feature detectors (SIFT, Harris corners), edge operators (Sobel, Canny) and many template-matching and OCR pipelines run on grayscale: it cuts memory and compute by three, removes color as a nuisance variable, and is often sufficient for the geometric task. The trade-off is that any cue that requires color — distinguishing a ripe red fruit from green leaves, or segmenting by hue — is irrecoverably lost, so grayscale is a deliberate choice, not a free simplification.