Ranking with uncertain labels and its applications

1 The techniques for image analysis and classi cation generally consider the image sample labels xed and without uncertainties. The rank regression problem is studied in this paper based on the training samples with uncertain labels, which is often the case for the manually estimated image labels. First, the core ranking model is designed as the bilinear fusing of multiple candidate kernels. Then, the parameters for feature selection and kernel selection are simultaneously learned by maximum a posteriori for given samples and uncertain labels. The convergency provable Expectation Maximization (EM) method is used for inferring these parameters in an iterative manner. The effectiveness of the proposed algorithm is nally validated by the extensive experiments on age ranking task and human tracking task. The popular FG-NET and the large scale Yamaha aging database are used for the age estimation experiments, and our algorithm signi cantly outperforms those state-of-the-art algorithms ever...

• Type: Journal
• Year: 2007
• Where: FCSC
• Authors: Shuicheng Yan, Huan Wang, Jianzhuang Liu, Xiaoou Tang, Thomas S. Huang
• Keywords: Convergency Provable Expectation, FCSC 2007, Human Tracking, Uncertain Labels