Fast reconstruction of corrupted images and videos by radial basis functions

This paper presents a novel approach for fast reconstruction of corrupted images and videos. It is expected that corrupted pixels are known and identified. The presented approach is based on a radial basis function interpolation from neighbouring uncorrupted pixels. The proposed algorithm is simple, easy to implement and extensible to higher dimension as well. Algorithm speed-up is mostly due to the inverse RBF matrices, dynamic memory management or GPU cache. An alternative implementation can be based on CPU multithreading giving, according to experiments, similar results as GPU based implementation.