<title>Modifications to Retinex to relax RESET nonlinearity and implement segmentation constraints</title>

Abstract

This paper addresses the Frankle-McCann Retinex algorithm by altering its properties to provide a distance-weighting function to every ratio-product path from a source to a destination pixel. The algorithm is further modified to permit the hard RESET function to be replaced by a piece-wise smoother function that allows ratio-product propagation to slightly exceed the maximum brightness in each visual channel. Investigations of how segmentation can aid in reducing the computational complexity and provide a more realistic white balance are presented.