Facial Expression Recognition Based on Genetic Programming Learning CCA Fusion
Min Wu,Ming Li,Chao He,Hao Chen,Yan Wang,Zhengxiu Li
DOI: https://doi.org/10.1109/prai55851.2022.9904275
2022-01-01
Abstract:The quality of the features directly affect the overestimation of image classification. However, it is difficult to distinguish between two categories only by a single feature when there is no obvious difference between them. One of the ideas to solve this problem is multi-feature fusion. However, the existing fusion methods have the problems of complex models, fixed models and the need for relevant knowledge in the field. Genetic programming (GP) is a feature fusion method with flexible representation. It can automatically learn other excellent fusion methods without operating how the computer fuses. According to this property, this paper presents a new LCGP method, which can automatically learn existing fusion methods. In the proposed method, the excellent fusion model will be refined into a mathematical form of the function operator. In fact, this is a three-tiered GP tree that integrates different features and fusion methods into the same tree through operators. The proposed method can automatically learn and evolve different mathematical fusion models, validate them on two small sample datasets, CK+ and JAFFE, and compare them with several state-of-the-art methods. The results show that although the training examples are limited, the performance of this method is better than or similar to that of the related methods.