A Model of Target Detection in Variegated Natural Scene based on Visual Attention

Less of edge and texture information existed in traditional visual attention model in target detection due to extract only the color, brightness, directional characteristics, as well as direct sum fusion rule ignoring the difference in each characteristic. A improved model is proposed by introduced the edge, texture and the weights in fusion rules in visual computing model. First of all, DOG is employed in extracting the edge information on the basis of obtained brightness feature with multi-scale pyramid using the ITTI visual computing model; the second, the non-linear classification is processing in the six parameters of the mean and standard deviation of the gray contrast, relativity and entropy based on the GLCM; finally, the fusion rule of global enhancement is employed for combination of multi-feature saliency maps. The comparison experimental results on variegated natural scene display, relative to the ITTI calculation model, there is more effective with the application of the model in this paper, the interested area and the order to shift the focus are more in line with the human visual perception, the ability of target detection is strengthening in variegated natural scene. Further shows that the proposed edge and texture features introduced in the primary visual features to be effective, the introduction of each feature significant weighting factor is reasonable in the feature map integration phase.