GRASS: A Gradient-Based Random Sampling Scheme for Milano Retinex

Abstract

Retinex is an early and famous theory attempting to estimate the human color sensation derived from an observed scene. When applied to a digital image, the original implementation of retinex estimates the color sensation by modifying the pixels channel intensities with respect to a local reference white, selected from a set of random paths. The spatial search of the local reference white influences the final estimation. The recent algorithm energy-driven termite retinex (ETR), as well as its predecessor termite retinex, has introduced a new path-based image aware sampling scheme, where the paths depend on local visual properties of the input image. Precisely, the ETR paths transit over pixels with high gradient magnitude that have been proved to be important for the formation of color sensation. Such a sampling method enables the visit of image portions effectively relevant to the estimation of the color sensation, while it reduces the analysis of pixels with less essential and/or redundant data, i.e., the flat image regions. While the ETR sampling scheme is very efficacious in detecting image pixels salient for the color sensation, its computational complexity can be a limit. In this paper, we present a novel Gradient-based RAndom Sampling Scheme that inherits from ETR the image aware sampling principles, but has a lower computational complexity, while similar performance. Moreover, the new sampling scheme can be interpreted both as a path-based scanning and a 2D sampling.

Using pixel intensity as a self-regulating threshold for deterministic image sampling in Milano Retinex: the T-Rex algorithm

“…Here we have chosen the measures f i 's, because they have been employed to evaluate Milano-Retinex algorithms previously published. 8,9,13,14 The image enhancement performance of T-Rex has been evaluated and compared with that of GREAT, RSR-P, and QBRIX on the dataset TEST35COLOR that has been already used to test previous Milano Retinexes. 9,14 This dataset consists of 35 real-world color pictures, taken under different light conditions, in indoor and outdoor environments.…”

“…Several sampling figures are proposed: for instance, random paths, 5,6 constrained random paths, [7][8][9] random sprays, [10][11][12] and edges picked up from random squares around the target. 13 In these cases, the pixelwise random sampling is repeated many times, and the lightness is computed by averaging the results obtained over each sampled set. These operations reduce the chromatic noise possibly affecting the output image due to the random sampling.…”

“…This undesired noise is completely avoided by the Milano Retinex approach GREAT, 14 which computes the spatial and color features relevant to the lightness from a set of strong, deterministically defined edges. With the exception of the methods based on random paths 5,6 that use a computational scheme similar to that of the original Retinex algorithm, the other methods mentioned above compute the local reference white by averaging the maxima of the intensity [7][8][9][10][11][12] or the maximals of the target intensity 13,14 over the sampled sets in this framework, the maximals of a real-value v over a set S ⊆ R are the elements of S strictly greater than v. If no maxima or maximals exist, the local reference white equals the target intensity. In some implementations, the contributions of the maxima or of the maximals are weighted by function of the spatial distance between the target and the pixels realizing the maxima or the maximals.…”

“…The main novelty of T-Rex is the pixelwise mechanism of self-regulating intensity thresholding. This mechanism makes T-Rex very different from Milano Retinexes based on random image exploration: [5][6][7][8][9][10][11][12][13] in fact, T-Rex implements a determistic image sampling, thus it avoids the generation of chromatic noise and allows the exact reproducibility of the results. T-Rex shares this characteristic with GREAT.…”

Gradient attenuation as an emergent property of reset-based Retinex models

“…In the original Retinex [20] and the Milano Retinex family [22,28] (a detailed description of their di erences can be found in [19,27]), the information from the surrounding is collected by statistical sampling -the samples are typically paths [14][15][16][17] or sprays [1,23]) -then sample-wise compared to the center/target pixel using an operator called reset, which in the simplest cases corresponds to the "max" operator (more details in the following sections). The reset outputs a value, which represents the contribution of that particular sample.…”