Adaptive visual saliency detection method via Hilbert-Huang Spectral Analysis

Visual saliency is largely determined by bottom-up factors that highlight image regions which are different from their surroundings. Based on the philosophy that exploits image information content as the metric of visual saliency, an adaptive visual saliency detection method(AVSDM) including Pixel Cluster, multi-scale Gaussian Pyramid Decomposition and Hilbert-Huang Spectral Analysis is proposed to measure the bottom-up factors in this paper. Saliency map is detected via non-redundant information based on shannon entropy theory. The method has a good performance in adaptively detecting tiny details as well as objects with complex background. Experiment results indicate that the proposed approach outperforms the state-of-the-art detection algorithms.