Unsupervised Human Action Categorization with Consensus Information Bottleneck Method

Recent researches have shown consensus clustering can enhance the accuracy of human action categorization models by combining multiple clusterings, which can be obtained from various types of local descriptors, such as HOG, HOF and MBH. However, consensus clustering yields final clustering without access to the underlying feature representations of the human action data, which always makes the final partition limited to the quality of existing basic clusterings. To solve this problem, we present a novel and effective Consensus Information Bottleneck (CIB) method for unsupervised human action categorization. CIB is capable of learning action categories from feature variable and auxiliary clusterings simultaneously. Specifically, by performing Maximization of Mutual Information (MMI), CIB maximally preserves the information between feature variable and existing auxiliary clusterings. Moreover, to solve MMI optimization, a sequential solution is proposed to update data partition. Extensive experiments on five realistic human action data sets show that CIB can consistently and significantly beat other state-of-the-art consensus and multi-view clustering methods.