Interactive Object Segmentation for mono and stereo applications: Geodesic Prior Induced Graph Cut Energy minimization

摘要

This study proposes an interactive multi label object segmentation method and applications on mono and stereo images. The general segmentation problem is approached by an energy minimization on a Markov Random Field (MRF). The minimum energy potential labelling is the primary goal of the multi label segmentation algorithm. User inputs are used to determine object location and geodesic prior induced iterative graph cut energy minimization is used to define object boundaries. Segmented objects on mono images are used to generate stereo pairs for viewing on 3D displays. Segmented object pairs on stereo images are used for depth adjustment in order to achieve better visual quality. The assignment of relative depths on multiple objects is necessary for stereo image pair synthesis using conventional depth image based rendering (DIBR) techniques.