Abstract

There is an emerging interest on using high-dimensionaldatasets beyond 2D images in saliency detection. Examplesinclude 3D data based on stereo matching and Kinect sen-sors and more recently 4D light field data. However, thesetechniques adopt very different solution frameworks, in bothtype of features and procedures on using them. In this paper,we present a unified saliency detection framework for han-dling heterogenous types of input data. Our approach build-s dictionaries using data-specific features. Specifically, wefirst select a group of potential foreground superpixels to build a primitive saliency dictionary. We then prune the out-liers in the dictionary and test on the remaining superpixels to iteratively refine the dictionary. Comprehensive experiments show that our approach universally outperforms the state-of-the-art solution on all 2D, 3D and 4D data.