Weighted peak ratio for estimating stereo confidence level using color similarity

Abstract

In this paper, we propose a new stereo confidence metric, weighted peak ratio. Unlike existing confidence metrics, it computes the confidence level using the costs of surrounding pixels and given weights based on the color similarity between the pixels. In the experimental results, the proposed confidence metric showed better performance in detecting outliers compared to the state-of-the-art confidence metric, average peak ratio. Especially, the proposed metric is effective in object boundary regions.