Principle:Kornia Kornia Grayscale Conversion
| Knowledge Sources | |
|---|---|
| Domains | Vision, Color_Space |
| Last Updated | 2026-02-09 15:00 GMT |
Overview
Technique of converting multi-channel color images to single-channel grayscale representations for feature extraction and matching.
Description
Grayscale conversion reduces a 3-channel RGB image to a single luminance channel by computing a weighted sum of the color channels. The standard ITU-R BT.601 weights (R: 0.299, G: 0.587, B: 0.114) reflect human perceptual sensitivity to different wavelengths. Grayscale images are required by many feature detectors (LoFTR, SIFT, HardNet) that operate on intensity gradients rather than color information. The conversion preserves spatial structure while reducing computational cost by 3x.
Usage
Use as a preprocessing step before feature detection, edge detection, or any operation that works on image intensity rather than color. Required by LoFTR and most classical feature detectors.
Theoretical Basis
gray = 0.299*R + 0.587*G + 0.114*B
In tensor form: gray = rgb_weights @ image, where rgb_weights = [0.299, 0.587, 0.114]. Custom weights can be provided for specialized applications.