A new method for saliency detection using top-down approach

Abstract

In this paper, we propose a visual saliency detection algorithm which used a learning method. In this model, we train a dictionary for twenty objects from Pascal VOC dataset and then we estimate saliency objects with project each image patch into the space of a dictionary of image patches (basis functions) learned from Pascal VOC dataset. We evaluate our method performance on two dataset along side state-of-the-art saliency detection methods and experimental results show that the proposed saliency model outperforms state-of-the-art saliency models.