۱- Introduction

۲- Related work

۳- Methodology

۴- Experimental results and analysis

۵- Conclusions

References

Abstract

Many effective methods have been proposed for face recognition in the past decade and the face recognition accuracy is also gradually improved, but these algorithms usually need to perform face alignment process based on the prior knowledge of facial structure before extracting facial features. The face recognition system usually consists of face detection, face alignment, facial feature extraction, etc., which are independent of each other, and it is difficult to design and train the end-to-end face recognition model. In this paper, an end-to-end face recognition method based on spatial transformation layer is proposed. Specifically, the spatial transformation layer is placed in front of the feature extraction layer of the face recognition network, and the face region is aligned by alignment learning which requires neither prior knowledge nor artificially defined geometric transformation. The face identity category information allows the convolutional neural network to automatically learn the most appropriate face alignment. Simulation experiments on CASIA-WebFace, LFW (Labeled Face in the Wild) and YTF (Youtube Face) face database have shown that the suggested alignment learning algorithm in this paper can realize the end-to-end face recognition and can effectively improve the face recognition rate as well.

Introduction

In the past decade years, the successful application of convolutional neural networks (CNN)in the image field has greatly improved the performance of computer vision tasks, such as face recognition and face verification [1, 2, 3]. The traditional face recognition methods are to construct the classification model mainly with artificially designed features, while the deep learning is to automatically learn more robust facial features through a large amount of training data. Therefore, CNN can achieve better recognition effects in the case of posture, occlusion, and illumination variation [4, 5]. However, the change of face posture is still one challenge of face recognition system in practical application. There are two ways to deal with such problems: one is to create a posture model to handle the facial posture change, Masi et al. [6] proposed an algorithm to calculate the posture distribution of the training data and establish two CNN models which correspond to the frontal face and the profile face respectively, and an excellent face recognition effect was obtained in the case of posture change. Liao et al. [7] suggested a partial face recognition localization method with multi-keypoint descriptors to represent align-free faces in which the descriptors’ size was determined by image content and face images. The other method is to introduce a face alignment process before facial feature extraction. Taigman et al. [8] developed a DeepFace network algorithm using deep learning for face recognition at the first time, and the 3D alignment method was used to solve the problem of out-of-plane rotations that traditional 2D alignment methods could not solve.