Author(s): MA Xiaolong, XIE Xudong, Lam Kinman, ZHONG Yisheng, Department of Automation, Tsinghua University, Department of Electronic and Information Engineering, the Hong Kong Polytechnic University
Pages: 1122-1126
Year: 2014 Issue:  8
Journal: Journal of Tsinghua University(Science and Technology)
Keyword:  saliency detection; robust principal component analysis; multiple color channels;
Abstract: Saliency detection is widely used in image segmentation,object detection and visual performance evaluations. Image preprocessing is enhanced by imitating the human visual mechanism with a saliency detection method,based on a robust principal component analysis algorithm and multiple color channels.The original image is first transformed into multiple color channels,represented by a matrix with the columns of this matrix linearly correlated.The salient regions are assumed to be the sparse component with the background regions as the low rank component.The robust principal component analysis of this matrix is used to extract the components.Use of a saliency prior and a center prior make the saliency detection model more effective.Tests show that this algorithm outperforms many state-of-the-art methods in terms of aquantitative index and the visual effect.