Using chromatic information in edge-based stereo correspondence

Abstract

One approach to developing a faster, more robust solution to the stereo correspondence problem is to seek a more complete and efficient use of available image information. Motivated by the observation that chromatic (color) information is a salient, regional property of surfaces in many natural scenes, the goal of this research has been to gain a fundamental understanding of the use of chromatic information in edge-based stereo correspondence. In particular, the use of chromatic information to characterize intensity edges is analyzed, and the chromatic gradient matching constraint, which specifies disambiguation criteria for edge-based stereo correspondence, is developed. The result is a theoretical construction for developing edge-based stereo correspondence algorithms which use chromatic information. The efficacy of using chromatic information via this construction is demonstrated by implementing a disparity-gradient-based algorithm and comparing algorithm performance with and without the chromatic gradient matching constraint. The results demonstrate that the use of chromatic information can significantly reduce the ambiguity between potential matches, resulting in increased algorithm accuracy as well as increased algorithm speed.