Saliency Detection Integrated with Depth Information

摘要

Most bottom-up saliency detection methods only perform outstandingly on simple image but when operating on complex one, results are not satisfactory. In a complex scene image, more image features demand being taking into consideration, the most important one of which is the spatial relation of different objects. Aiming at this problem, a new image saliency detection method is proposed based on frequency-tuned (FT) salient region detection. In our method, depth map of the image are constructed and then depth information of every object are added to the calculation of saliency to improve FT. Experimental results illustrate that our proposed method outperforms the original one and thus the salient regions are more evident in the ultimate saliency map.