Salient object detection on multiscale learning and sparse coding
Ying Tang, Lei Zhu, Jin Wu · 2017
In this paper, we introduce a saliency detection method to automatically detect salient regions of different kinds of images with different complex scenes. Our method takes the advantages of efficiency and robustness of machine learning and sparse coding. We adopt concept of multiscale learning to use random forests classifier to get the training model. Then we construct non-saliency dictionary through the potential background information of initial map produced by using the training model. To refine the saliency map and achieve better saliency performance, we utilize weighted sparse coding to compute the saliency map with the non-saliency dictionary which contains the most foreground information. The experiment results indicate that our method is intuitive, effective and achieves state-of-the-art results on several benchmarks.