A Study of Top-Down Visual Attention Model Based on Similarity Distance

Abstract

According to the study of biological vision, human visual attention is aroused by observer's tasks and it is a top-down control mechanism. A top-down visual attention model based on similarity distance is proposed in this study. Top-down visual attention model is realized by using similarity distance to obtain visual expectation based on improved Itti's model. A task map is selected by using improved Itti's model and user's observation tasks, and then to compute the similarity distances between task map and saliency regions from improved Itti's model. The weights of intensity, color and orientation are adjusted with similar distances in order to measure the saliency in top-down visual attention model. After that, area enhancement factor is utilized to modify the saliency and compute the visual expectation. Region of interest (ROI) is extracted by combining visual expectation with saliency region. The experimental results show that the proposed model is available and the ROI results comply with biological attention mechanism.