Face Recognition Based on Local Gabor Binary Patterns and Convolutional Neural Network

Abstract

Enhancing the robustness to changes caused by facial aging in automatic face recognition system is still an important problem worth researching. Compared with the external factors, such as illumination, posture and expression, facial aging which can produce variations in both shape and texture of the face has more complex effects. In this paper, we propose a method based on Local Gabor Binary Patterns and Convolutional Neural Network (LGBP-CNN) to improve the performance of age invariant face recognition problem. For each face image, this method first extracts shape, texture and local neighbor relationship features with multi-orientation and multi-scale Gabor filters as well as local binary patterns (LBP) operators. Then, we utilize one kind of Deep Learning model-convolutional neural network which has shown brilliant performance on face recognition area to avoid the dimension curse problem brought by Gabor filtering and further extract features. Such kind of method has robustness to changes of illumination, posture, expression, shape and texture by combining Gabor transform, LBP and convolutional neural network. Experiments are implemented on the FG-NET database and the results can outperform the state of the art ones, which verify the validity of the proposed method in age invariant face recognition problem.