Discriminative metric learning for face verification using enhanced Siamese neural network

  • PDF / 1,059,010 Bytes
  • 18 Pages / 439.37 x 666.142 pts Page_size
  • 54 Downloads / 248 Views

DOWNLOAD

REPORT


Discriminative metric learning for face verification using enhanced Siamese neural network Tao Lu 1

1

1

& Qiang Zhou & Wenhua Fang & Yanduo Zhang

1

Received: 16 April 2020 / Revised: 21 August 2020 / Accepted: 28 August 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract

Although face verification algorithms have made great success under controlled conditions in recent years, there’s plenty of room at its performance under uncontrolled realworld due to lack of discriminative feature representation ability. From the perspective of metric learning, we proposed a context-aware based Siamese neural network (CASNN) to learn a simple yet powerful network for face verification task to enhance its discriminative feature representation ability. Firstly, a context-aware module is used to automatically focus on the key area of the input facial images without irrelevant background area. Then we design a Siamese network equipped with center-classification loss to compress intraclass features and enlarge between-class ones for discriminative metric learning. Finally, we propose a quantitative indicator named “D-score” to show the discriminative representation ability of the learnt features from different methods. The extensive experiments are conducted on LFW dataset, YouTube Face dataset (YTF) and real-world dataset. The results confirm that CASNN outperforms some state-of-the-art deep learning-based face verification methods. Keywords Metric learning . Discriminative feature . Siamese neural network . Face verification

1 Introduction Due to its nonintrusive and natural characteristics, face verification, which is a prominent biometric technique for identity authentication, has been widely used in many areas, such as military, finance, public security even daily life [27]. Face verification method mainly includes three steps: extracting features from the input two face images, calculating similarity, and judging by similarity whether these two face images are from the same person or not. Thus, the

* Tao Lu [email protected]

1

Hubei Key Laboratory of Intelligent Robot, School of Computer Science and Engineering, Wuhan Institute of Technology, Wuhan 430073, China

Multimedia Tools and Applications

performance of face verification algorithms is mainly determined by how to learn the discriminative features. From the perspective of feature extraction, face verification algorithms can be divided into three categories, geometric-feature based, local-feature based and global-feature based. In the early years, face verification algorithms mainly focused on matching the geometric features of different facial images. Due to the structure feature of facial images, early face verification methods attempted to distinguish the geometric distance of facial organs. Through the histogram projection of facial grayscale [2, 9, 12, 18] or template matching [8] to aid in determining the position of the facial organs, and extracting a series of geometric features related to the facial organs. Such as the spa