Attribute-restricted latent topic model for person re-identification

Searching for specific persons from surveillance videos captured by different cameras, known as person re-identification, is a key yet under-addressed challenge. Difficulties arise from the large variations of human appearance in different poses, and from the different camera views that may be involved, making low-level descriptor representation unreliable. In this paper, we propose a novel Attribute-Restricted Latent Topic Model (ARLTM) to encode targets into semantic topics. Compared to conventional topic models such as LDA and pLSI, ARLTM performs best by imposing semantic restrictions onto the generation of human specific attributes. We use MCMC EM for model learning. Experimental results show that our method achieves state-of-the-art performance.