Correspondence and line field estimation using MAP-based probabilistic diffusion algorithm

This paper proposes the dense correspondence field estimation with the stochastic diffusion based on maximum a posteriori (MAP) estimation. MAP-based correspondence field estimation including line field is derived with reflecting the joint probabilistic distribution function (PDF) of the neighborhoods in the Markov random field (MRF) models, and is applied to the stereoscopic images. The exploits of the joint PDF of the neighborhoods is the main difference from the previous MAP-based algorithms. The segmentation field is introduced in the low activity region, where the intensities are uniform, in order to improve the performance of the correspondence estimation. According to the experiments, the proposed algorithm had good estimation performance with fast convergence. Especially, the line field improved the estimation at the object boundaries, and the segmentation field did effectively in the low activity region