Using a small number of training instances in genetic programming for face image classification

Classifying faces is a difficult task due to image variations in illumination, occlusion, pose, expression, etc. Typically, it is challenging to build a generalised classifier when the training data is small, which can result in poor generalisation. This paper proposes a new approach for the classification of face images based on multi-objective genetic programming (MOGP). In MOGP, image descriptors that extract effective features are automatically evolved by optimising two different objectives at the same time: the accuracy and the distance measure. The distance measure is a new measure intended to enhance generalisation of learned features and/or classifiers. The performance of MOGP is evaluated on eight face datasets. The results show that MOGP significantly outperforms 17 competitive methods. (C) 2022 Elsevier Inc. All rights reserved.

Automated summary

This paper proposes a new approach for face image classification based on multi-objective genetic programming. It automatically evolves image descriptors that extract effective features by optimizing both accuracy and distance measure, aiming to enhance generalization. Experimental results on multiple datasets demonstrate that this method significantly outperforms other competitive methods.