Unsupervised learning of visual context using completed likelihood AIC
Tao Xiang · 2005
Learning visual context is a critical step of dynamic scene modelling. This paper addresses the problem of choosing the most suitable probabilistic model selection criterion for learning visual context of a dynamic scene. A Completed Likelihood Akaike’s Information Criterion (CL-AIC) is formulated to estimate the optimal model order (complexity) for a given visual scene. CL-AIC is designed to overcome poor model selection by existing popular criteria when the data sample size varies from very small to large. Extensive experiments on learning visual context for dynamic scene modelling are carried out to demonstrate the effectiveness of CL-AIC, compared to that of BIC, AIC and ICL.