RGB-D geometric features extraction and edge-based scene-SIRFS

Abstract

In this paper, we propose a geometric features extraction approach for RGB-D image and its application in intrinsic scene properties recovery. Our approach extracts geometric features, and make use of both 3D shape information from depth map and edge information from RGB image. Geometric features are categorized as occlusion edges and fold edges. The proposed method can not only extract the edge information from the RGB-D image accurately and efficiently, but also classify the edge information effectively. Moreover, our algorithm is close to the real time processing. Besides, we present an application of these geometric features as edge-based intrinsic scene properties recovery. Experimental results demonstrate that our proposed of geometric features extraction and edge-based intrinsic scene properties recovery outperforms Scene-SIRSF algorithms.