Study of target detection based on top-down visual attention

This paper has presented a three-stage method which combines the visual attention model with target detection. This method thresholds the saliency map, covers the region of interest with blocks and measure similarity between the blocks and the template. It is a new approach to segment an image on the domain of saliency to select the region of interest which reduce the number of translations that need to be checked in large amount. Kernel density approximation method is introduced, which provides a flexible and accurate framework to represent the template. The result of the test designed in this paper shows the feasibility of the method.