확률적 확산 모델을 이용한 스테레오 정합 및 객체추출

Abstract

In the paper, srochastic models are proposed for simultaneous correspondence (motion or disparity) estimation an object segmentaion. The correspondence and segment fields are explicitly modeled as markov random fields, and estimated in the energy minimization process called stochastic diffusion. The potential spaces of the two fields are stochastically diffused with the probability distributions of the neighboring fields, andthe fidls are estimated from the atabilized potential spaces. The stochastic diffusion as an energy minimizaion process is a kind of generative model shice updates and regenerates the probability spaces of the correspondence and segment fields. And, some experiments are perfomed on the simultaneous correspondence esrimation and object segmentation for the stereocopic images and video sequences.