A study on statistical feature extraction from three-dimensional spatio-temporal data

点击翻译按钮获取中文摘要

In this study, we have proposed novel methods to extract features from spatio-temporal data for motion/image classifications. Those methods are based on co-occurrences of primitive descriptors in local neighborhoods in the spatio-temporal domain. By employing as the descriptors the space-time gradients and local feature vectors, we obtain the effective features to describe motions and images. In the experiments on motion/image classifications, the proposed methods exhibited favorable performances.