Fast gradient histogram specification for suppression of radiometric differences of single sensor stereo image pair

Abstract

Cross-channel depth sensing camera systems, such as Offset Aperture (OA) camera, employ an NCC based correspondence measurement, as it is more robust against certain radiometric differences. However, the NCC is a complex matching cost and results in depth blurring. In this paper, we propose a computationally simpler depth extraction scheme, where the radiometric differences of the OA stereo images are suppressed using the gradient histogram specification and an SAD based correspondence measurement is performed. We also propose a fast gradient histogram specification (FGHS) approach, where the gradient cumulative distribution function (CDF) is estimated using local signal statistics as Cauchy distribution. Proposed FGHS approach effectively suppresses the radiometric differences of the stereo image gradients compared to classical histogram specification (HS) and exact histogram specification (EHS) approaches. Furthermore, proposed FGHS-SAD based depth extraction scheme provides fast, and more accurate depth results compared to the current NCC based OA depth extraction scheme as well as HS-SAD and EHS-SAD based depth extraction schemes.