Top-Down Models Of Human Visiual Attention Using Dynamic Bayesian Network

Abstract: Visual attention is one of the built in mechanisms in system that quickly selects regions in a visual scene, which are most likely to contain items of interest. Many computational models are developed to predict the behaviour of human visual attention. This paper describes the computational model involving top-down process for most attended region using Dynamic Bayesian Network. By using Dynamic Bayesian Network, Top-down knowledge is predicted. Top down saliency map is found using three different methods namely weighted modulation, weighted combination and joint learning. The performance of each method is compared using Normalized scan-path saliency (NSS).

Keywords: Attention, Dynamic Bayesian Network, Top-Down Knowledge