| Summary: | This paper proposes an effective method to improve the saliency detection performance of existing RGBD (RGB image with Depth map) saliency models. First, a progressive region classification method is proposed to collect training samples at coarse scale and fine scale via the inter-region hierarchical structure. A random forest regressor is then learned to predict the coarse saliency map and fine saliency map, respectively. Finally, the saliency maps at the two scales are integrated into the final saliency map under the constraint of the inter-region hierarchical structure. Experimental results on a RGBD image data set and a stereoscopic image data set with comparisons with the state-of-the-art saliency models validate that the proposed method consistently improves the saliency detection performance of various saliency models.
|
|---|
