广义数据场及其在人脸表情识别中的应用

In this paper, a generalized data field is proposed to better estimate the potential function in multi-dimensional data field, along with a case study on face recognition. Data field is a method to understand and depict the data characteristics in the context of physics. In view of the inadequacy of the existent potential function, the data field is further developed to become the generalized data field by extending the impact factor from isotropy to anisotropy in multidimensional space. Focusing on the accuracy of potential function estimation, the performance of potential function estimation is primarily determined by impact factors and only in a minor way by unit potential function. Facial expressions are various, complicated and changeable, and they are actually caused by the changing interaction among different partitions in a face. Thus a case study on face recognition is experimented using JAFFE database. The whole recognition rate is up to 94.3%. The final results show that the generalized data field is valid for face recognition.