Adaptive cross-trilateral depth map filtering

Abstract

Stereo-correspondence algorithms often produce noisy and inaccurate depth estimates. We present a mechanism for improving depth maps by means of adaptive cross-trilateral median filtering (ACTMF). To preserve details in the depth map and simultaneously smooth depths in homogeneous areas the filter parameters are adapted to the local image structure. The influence of initial miss-matches is minimized by the introduction of an additional confidence kernel that penalizes unreliable matches. Experimental results show a significant improvement of the initial input depth maps.