Composition of Similarity Metrics for Correspondence Matching in Depth Estimation

Correspondence matching is a prerequisite step in dense depth estimation techniques. In this paper we consider various similarity metrics for correspondence matching and we present an approach which can be used to optimize it. Experimental results show that by careful selection of similarity metric can have positive impact on depth estimation quality and that the differences between various metrics range up to 60 percent points of bad-pixel depth map quality ratio. It has also been shown that usage of proposed composite similarity can lead to improved depth map quality, expressed as lower bad-pixel ratio.