Visual saliency detection based on adaptive fusion of color and texture features

摘要

Visual saliency detection achieved excellent performance in various fields such as object detection, image compression, and image retrieval. However, most existing methods for visual saliency detection only considered low-level features, ignored higher-level priors, and the fusion mechanism was simple. A novel visual saliency detection model based on color and texture adaptive fusion was proposed in this paper. On the basis of image preprocessing, the proposed method extracted color saliency map through color contrast feature and color distribution feature fusion, and texture saliency map through texture feature. Then they were fused adaptively according to the texture complexity of each image. The final saliency map was obtained by incorporating location prior. Experimental results on MSRA (1000) dataset demonstrated that the proposed visual saliency detection model outperformed the existing methods.