Saliency detection using superpixel belief propagation

We propose a method to detect saliency from a single image using feature extraction and superpixel belief propagation. We observe that the previous works are hard to deal with the intrinsic material discontinuity and non-homogeneous color distribution within an object or a region. Motivated by this observation, we bring the belief propagation into the saliency detection. First, we separate the image into middle-level superpixels and also extract the low-level feature within each superpixel. Then, we build up a Markov-Random-Field (MRF) on the middle-level super-pixels and adopt propagation technique to optimize the superpixel saliency. Afterward, we refine this middle-level solution to per-pixel saliency map. Experimental results demonstrate that our proposed method is promising as compared to the state-of-the-art methods in both MSRA-1000 and SED datasets.