Asymmetric depth estimation for multiview

A novel multi-view depth map estimation algorithm is proposed with low computational complexity and high accuracy, while meeting requirements of N-view plus N-depth representation. The algorithm exploits an aggregation methodology, namely permeability filtering (PF), through computationally efficient two-pass aggregation producing adaptively weighted and connected 2D support regions for each pixel. This aggregation approach is quite efficient compared to its alternatives in the literature. The advantage of being occlusion free in multiple viewpoints results with high quality maps compared to stereo matching. Moreover, the asymmetric approach further increases efficiency without any loss in the precision, in terms of mapped reliable center depth map to neighboring cameras in order to avoid redundant computation. Hence, the proposed algorithm is shown as to be a competitive alternative via simulations to some leading state-of-the-art methods by reduced number of computations and efficient support calculation for extracting high quality dense depth maps from multiple camera views.