Statistical Modeling Using Bounded Asymmetric Gaussian Mixtures: Application to Human Action and Gender Recognition

To determine the structure of high dimensional data without knowing the number of clusters nor the importance of the involved features, we propose an unsupervised feature selection framework using the bounded asymmetric Gaussian mixture model (BAGMM-FS). The bounded asymmetric Gaussian distribution has an asymmetric shape and bounded range, making it a good choice for modeling real-world data. We propose a parameter learning approach based on the expectation-maximization (EM) algorithm, and we approach the model selection task using the minimum message length (MML) criterion. The validation involves several human-related recognition challenges, such as human activity categorization and human gender recognition. It's examined from all experiments and results that BAGMM-FS has good modeling capabilities and outperforms other comparable mixture models, especially for high dimensional complex datasets.