3-D Layered Representation of Multiview Images Based on 3-D Segmentation

Abstract

In this paper, we propose a new approach to represent a set of multiview images with surface models corresponding to objects in the scene. Because the surface models overlap each other, the proposed method can cope with complex scenes in which occlusion occurs. First, a set of multiview images is partitioned into 3-D regions corresponding to the scene objects. Second, the structure of each 3-D object cor-responding to a 3-D region is recovered and represented by a surface model using triangular patches. These surface models are then placed in appropriate locations to describe occlusions. This method can be seen as an extended version of layered representation. Each layer is an uneven surface, while previous works have concentrated on planar layers. The experimental results show how the 3-D layered representation method could be applied to the next-generation 3-D image communication system.