Unsupervised Mask Patterns Generation for Extracting Action Specific Motion Features

This paper presents unsupervised mask patterns generation for extracting action specific motion features. Cubic Higher-order Local Auto-Correlation (CHLAC) feature is robust to position changes of human actions in a video, and it is effective for action recognition. However, the mask patterns for extracting features are fixed. In other words, the mask patterns are independent of action classes. This is a merit but the features extracted from those mask patterns are not specialized for each action. Thus, we make mask patterns automatically for extracting action specific features by clustering of local spatio-temporal regions in each action. Since how to extract features by the proposed mask patterns is the same as CHLAC, our method also has shift invariance property. By the experiments using the KTH dataset, the effectiveness of our method is shown.