Natural Color Satellite Image Mosaicking Using Quadratic Programming in Decorrelated Color Space

Abstract

Generating mosaics of orthorectified remote sensing images is a challenging task because of the colorimetric differences between adjacent images introduced by land use, surface illumination, atmospheric conditions, and sensor. Most of the existing color correction methods involve pairwise techniques, which are limited when the collection of images is large with numerous overlaps. Besides, available techniques do not operate in a color space suited for true-color processing. This paper presents a simple and robust method to perform the global colorimetric harmonization of multiple overlapping remote sensing images in natural colors (RGB). Our parameter-free method deals simultaneously with any number of images, with any spatial layout, and without any single reference image. It is based on the resolution of a quadratic programming (QP) optimization problem. It operates in the lαβ decorrelated color space, which is well suited for human vision of natural scenes. The results obtained from the mosaicking of 132 RapidEye color orthoimages over mainland France demonstrate good potential for performing colorimetric harmonization automatically and effectively.