Salient Object Detection Via Multi-layer Markov Chain
Salient object detection is a fundamental computer vision task, which aims to identify and locate the most interesting objects attracting visual attention of humans in an image.In this work, we regarded the problem of salient object detection as Multi-layer Markov random walks on an image graph mode.Superpixels are adopted to represent images as graphs.We computed the weights of the edges by a set of features including background prior, dissimilarity, and similarity.Here, we decided the state of each node of next layer Markov Chain by the states of its neighbors and the present corresponding state.For equilibrating the influential power towards the state of every node, we constructed the dissimilarity and similarity features.If a random walker were to walk eternally, the fraction of the time that he would spend at every node could be reflected by the equilibrium distribution of Markov chain.Hence,we used the equilibrium distribution as the indication of saliency.Four challenging datasets with different scenarios are adopted to evaluate our model for salient object detection.Our method performs better than the twenty state-of-the-art methods according to experimental results.
Saliency detection Markov Chain Graph Model Equilibrium Distribution
Lin Mingqiang Dai Houde Bao Hua Zeng Yadan
Quanzhou Institute of Equipment Manufacturing, Haixi Institutes, Chinese Academy of Sciences, Fujian School of Electrical Engineering and Automation, Anhui University, Anhui, Hefei, 230601
国内会议
第20届中国系统仿真技术及其应用学术年会(20th CCSSTA 2019)
合肥
英文
534-540
2019-08-01（万方平台首次上网日期，不代表论文的发表时间）