A linear manifold representation for color correction in digital images

Abstract

Images captured using a camera loses its dynamic range of colors as they are digitized. This problem is not encountered by the human visual system as it supports a wider dynamic range. Our enhancement model is based on the human visual system involving three processing steps-color characterization, color enhancement and color correction. Each pixel in an image, along with its neighborhood forms color manifolds in RGB space. In the proposed color characterization method, these manifolds are modeled as lines. In the color enhancement step, a hyperbolic tangent function compresses the dynamic range of the image. This nonlinear function enhances the image preserving its details, but not the color relationships. Each enhanced pixel is projected to a point on the best fit line corresponding to its manifold to restore the original color relationships. Being a single-step convergence algorithm, it is faster than other iterative methods. © Springer-Verlag 2011.