Mesh motion scale invariant feature and collaborative learning for visual recognition

摘要

Visual recognition has been gradually played important roles in many fields. An effective feature descriptor, with higher discrimination and higher descriptiveness for the different visual recognition tasks, is a challenging issue. In this paper, we propose a novel feature, called mesh motion scale invariant feature description, to facilitate the different visual task description and balance discrimination and efficiency. Then, a hierarchical collaborative feature learning model for multi-visual tasks in complex scenes is presented for obtaining the recognition results. Four large databases, FRGC, CASIA, BU-3DFE and 3D Online Action, are introduced to the performance comparison and the experimental results show a better performance for face recognition, expression recognition and activity recognition based on our proposed method.