Visual saliency detection based on visual center shift

Abstract

The saliency areas extracted by traditional visual saliency detection methods are not clear enough. This paper presents a visual saliency detection method based on visual center offset. On the basis of pre-segmentation of the image, the significant areas of the image are extracted by combining the color contrast, color distribution and location characteristics. Using visual center transfer to simulate the visual transfer process of human observation, the image is analyzed at multiple scales. The results indicate that this approach is efficient because ROC curve and Precision-Recall performed well.