A visual saliency model based on local entropy

Abstract

Human vision relies on visual attention mechanism to select the relevant parts of a scene, where higher level tasks could be completed. A model of selective attention is proposed based on local entropy. Firstly, visual features are extracted by independent component analysis. Then the DOG filter is applied to enhance the feature map. Local entropy transforms the feature map into the saliency map. The performance of the proposed model is evaluated by predicting human fixations in natural images. Experimental Results show that the model could achieve competitive performance. © 2011 Springer-Verlag.