A Unified Approach to Colour2grey and Image Enhancement Through Gradient Field Integration

TL;DRAbstract

The standard approach for the conversion of colour images to greyscale is to remove their chromatic content and keep only the brightness information. However, with this method the detail at equiluminant edges disappears. In this paper, we propose a colour-to-greyscale transformation that preserves details and, at the same time, enhances the output. We consider Socolinsky and Wolff's technique that codes the contrast information of the three colour channels in a single gradient field. Our contribution is to show how retinex-type algorithms produce a greyscale image by integrating such a gradient. Our approach also addresses the question of integrability where the gradient field is non-integrable (causing the greyscale to have smears and bending artefacts). We show how by diminishing the saturation of the original input we not only reduce the integrability problem but, with enhancement, keep the perceived colour contrast in a greyscale reproduction.

Chat with Paper

AI Agents for this Paper