Efficient aggregation via iterative block-based adapting support-weights

Local stereo matching algorithms based on adapting-weights aggregation produce excellent results compared to other local methods. In particular, they produce more accurate results near disparity edges. This improvement is obtained thanks to the fact that the support for each pixel is accurately determined based on information such as colour or spatial distance. However, the computation of the support for each pixel results in computationally complex algorithms, especially when using large aggregation windows. Iterative aggregation schemes are a potential alternative to using large windows. In this paper we propose a novel iterative approach for adapting-weights aggregation which produces better results and out-performs most previous adapting-weights methods.