Improving Phase-Based Disparity Estimation by Means of Filter Tuning Techniques

Phase differencing techniques have been proven to be fast and robust methods for estimating disparity between two views. This disparity estimation depends on the quality of the local phase information which is a response of carefully designed frequency selective filter pairs for local phase estimation. Badly adjusted filter kernels yield responses with low amplitude and thus numerically instable phase information. In this paper we investigate the role of filter tuning to avoid singular points. We present a new iterative algorithm to optimally adjust the local phase estimating filters and compare the results with other phase differencing techniques as well as an instantaneous frequency driven filter tuning. Various experiments demonstrate that the iterative filter tuning technique shows improved performance.