Object detection based on saliency map
Dan Xu · Journal of Computer Applications · 2010
This paper presented a method which could detect salient object based on saliency map.This method added curvature to bottom-up visual attention model in order to obtain more approximate shape of the object.Firstly,intensity,color and orientation feature maps were extracted by bottom-up visual attention model.Total three normalized feature maps obtained by center-surround approach were linearly combined to saliency map.Curvature saliency map was processed to highlight more salient shape and reduce the negative effect of the surrounding pixels by region-grow segmentation algorithm.Salient object location was obtained by effective sub-window search in saliency map which was processed by threshold.Experimental results show the effectiveness of this method.