Novel Trilateral Approach for Depth Map Spatial Filtering

Abstract

In this paper, we present our approach for spatial filtering of depth map extracted from camera motion. An original depth map may have some artifacts owing to imperfect motion estimation. Our goal was to make the depth map uniform in smooth areas and to refine object boundaries without blurring edges. To solve this problem we propose the trilateral filter, whose convolution kernel is composed of a distance kernel, a color-based kernel and a depthbased kernel. Experiments demonstrate that this approach yields rather good results. Also, we compare our results with those of a typical bilateral filter.