STAR: A SEGMENTATION-BASED APPROXIMATION OF POINT-BASED SAMPLING MILANO RETINEX FOR COLOR IMAGE ENHANCEMENT

Abstract

Milano Retinex is a family of spatial color algorithms inspired by Retinex and mainly devoted to the image enhancement. In the so-called point-based sampling Milano Retinex algorithms, this task is accomplished by processing the color of each image pixel based on a set of colors sampled in its surround. This paper presents STAR, a Segmentation based Approximation of the point-based sampling Milano Retinex approaches: it replaces the pixel-wise image sampling by a novel, computationally efficient procedure that detects once for all the color and spatial information relevant to image enhancement from clusters of pixels output by a segmentation. The experiments reported here show that STAR performs similarly to previous point-based sampling Milano Retinex approaches, and that STAR enhancement improves the accuracy of the well-known algorithm SIFT on the description and matching of pictures captured under difficult light conditions.

Existing System

Light Energy-driven Termite Retinex (L-ETR), Termite Retinex (TR) algorithm

Proposed System

This paper presents STAR, a novel spatial color algorithm for image enhancement, inspired by the Milano Retinex approaches. STAR is a Segmentation based Approximation of Milano Retinex sampling based methods, which aims at cutting down the computational burden of the image sampling and at reducing the number of operations needed to compute the lightness. STAR accomplishes this task by employing coarse color and distance information, computed from clusters of pixels detected by a segmentation and independent of the target. Moreover, STAR models locality by considering the mutual distances between the segments, instead of the pixelwise distances between the target and the sampled pixels, as usually done in many other Milano Retinex approaches. STAR takes as input a RGB image and processes its color channel independently. The lightness of each channel is computed by two steps. In the first step, STAR performs a global channel analysis. STAR segments the channel in many regions, let’s say P1…, PM, with M > 0. For each region Pj, STAR selects a set of representative elements, which are the pixels maximizing the distance transform over Pj, and computes the maximum intensity value over Pj. Then, it computes the mutual distances between the segmented regions as the minimum distances between their representative elements. In the second step, STAR implements a pixel-wise analysis. For each target x, STAR detects the segment Pi including x, then computes the lightness at x as the average of the maximals of the target intensity over the M intensity values selected from the channel segments. The contribution of every maximal to the lightness is weighted by a function inversely proportional to the distance between Pi and the region containing the maximal, defined in the previous step.

Architecture

BLOCK DIAGRAM