NEURAL SMOOTHERS FOR NOISED DYNAMIC IMAGE SEQUENCES

Rastislav Lukáč - Csaba Stupák - Stanislav Marchevský

   This paper is focused to exploit neural networks as smoothers for noised dynamic image sequences. For the learning capability and noise suppression property, the neural networks are very popular and widely used in smoothing applications. In addition, thanks to the development of VLSI technology, the implementation of neural networks is simple. In this paper, three approaches of networking are presented. Thus, the neural network as the temporal, spatial and spatiotemporal filter is proposed. The well-known and very popular backpropagation algorithm is used. The performance of neural smoothers for dynamic image sequence is evaluated through objective criteria and compared with traditional temporal, spatial and spatiotemporal medians.

Keywords: neural networks, image processing, backpropagation, smoother, median

[full-paper]