Adaptive FOA for saliency-based visual attention

摘要

This paper describes an adaptive extraction of focus of attention for saliency-based visual attention. On the contrary to the existing fixed-size FOA, we proposed a shape-based FOA according to the most salient region from saliency map of input image. We determine the most salient point by checking every value in saliency map, and expand the neighborhood of the point until the average value of the neighborhood is smaller than 75% value of the most salient point, and then find the contour of the neighborhood. Therefore our adaptive FOA is close to the shape of attended object and it is efficient to the object recognition or other computer vision fields.