An Occupancy-Depth Generative Model of Multi-view Images

This paper presents an occupancy based generative model of stereo and multi-view stereo images. In this model, the space is divided into empty and occupied regions. The depth of a pixel is naturally determined from the occupancy as the depth of the first occupied point in its viewing ray. The color of a pixel corresponds to the color of this 3D point. This model has two theoretical advantages. First, unlike other occupancy based models, it explicitly models the deterministic relationship between occupancy and depth and, thus, it correctly handles occlusions. Second, unlike depth based approaches, determining depth from the occupancy automatically ensures the coherence of the resulting depth maps. Experimental results computing the MAP of the model using message passing techniques are presented to show the applicability of the model.