Visual Saliency Detection Based On Visual Center Shift

Vision is the main way for people to obtain information and understand the world.It is very easy for human to recognize the image and judge the saliency regions,since the high parallelism of human visual system and the integrity which human eyes observe images.With the rapid development of information technology and digital media,visual saliency detection has been widely used in many fields,including remote sensing image,medical image processing,robot vision control,etc.The research and application of visual saliency detection technology make the computer process information efficiently and rapidly as same as human visual system,has an important role in improving image understanding and processing system,the practical application level of image processing technology.Because the existing saliency detection method can not extract saliency target accurately and efficiently,based on human visual attention,combining color contrast features,color distribution features and location features,fusing features in multi-channels,after multi-scale analysis,the saliency map was extracted.The main work includes the following aspects:(1)Since the image's own prior information is not used by the traditional salient detection method,a method is proposed which combines the prior information on the image background with the color features.Based on the features of image edge regional differences in the background are considered in the model,missing the objects in the edges of the image can be avoided.Then the two significant features are fused to obtain the final saliency map.(2)Visual saliency detection was widely used in the field of computer vision,but the saliency regions extracted by many detection methods were not clear enough.A novel saliency detection method based on visual center offsetting was proposed in this paper.On the basis of images' pre-segmentation,combining with color contrast features,color distribution features and location features,saliency region of an image was extracted.The center offsetting was used to simulate the vision transfer process of human,after multi-scale analysis,by fusing saliency maps at different scales,the final saliency map was computed.We made experiment on date sets of MSRA.The results showed that the proposed method performanced better on visual effect,the ROC curve and the precision recall rate than existing methods.