Edoardo Provenzi (Telecom ParisTech)

From histogram equalization to perceptually-inspired color enhancement: a variational approach

mardi 28 mai 2013, 13h45 - 14h45

Salle de réunion, espace Turing

{{Abstract :}}

The color histogram of an image is basically the distribution of its color values in the three chromatic channels after the specification of a color space. Considering the RGB space, the histogram equalization consists in equalizing the Red, Green and Blue channels separately, which leads, in general, to artifacts and noise amplification. In 1997 it has been proven that is possibile to interpret histogram equalization as the minimization of a suitable functional. This result allowed a better comprehension of histogram equalization of a digital image as the optimal balance among intensification of the global Michelson contrast and minimization of the quadratic dispersion around the middle gray. In the seminar it will be discussed how this variational framework can be considered as the starting point of a recent model for perceptually-inspired color enhancement of color images in the RGB space overcoming the problems of histogram equalization. In this model, the Michelson contrast must be replaced by a more suitable local and perceptually sound measure, while quadratic dispersion must be substituted by the entropy distance. Finally, it will be shown how to write and minimize these kind of functionals in the wavelet domain. In the first part of the seminar, the most technical concepts will be recapped

{{L’exposé sera donné en anglais.}}