Multi-angled Face Segmentation and Identification Using Limited Data

This paper introduces a different approach to face segmentation that aims to improve face recognition when given large pose angles and limited training data. Face segmentation is achieved by extracting landmarks which are manipulated in such a way as to normalize unseen data with a classification model. The approach is compared with related systems, followed by further tests that show consistent results across other datasets. Experiments include frontal and non-frontal training images for classification of various face pose angles. The proposed system is a promising contribution, and especially shows the importance of face segmentation. The results are achieved using minimal training data, such that both accurate and practical face recognition systems can be constructed.

I. Introduction

Since very early in civilization, human recognition has been a simple but restricted task. Human traits such as faces can be visually identified to ascertain familiar and unfamiliar individuals in a fairly unconscious manner [1]. However, the task has become increasingly challenging with the rise in populations. This is especially a concern with cross-border security and crime being pressing issues in the current era. While biometrics research has become more popular, it is reliant on using traits that are unique and easily distinguishable. Biometrics is the measurement and analysis of unique biological and behavioural traits to establish the identity of a human being for a particular purpose, often related to access control or law enforcement [2]. Face recognition research is important as it is regarded as one of the more visible and user-friendly biometrics. In cases where the face is not used as an authentication biometric, it is still often bound to the primary identification device or system, such as an ID card or a criminal fingerprint database [1]. As a primary means of identification, the face is challenging to segment in uncontrolled applications, such as cross-border security and surveillance, due to varied pose angles and occlusions. Furthermore, real-world conditions often result in degradation of the quality when acquiring face images from camera sensors. When bound as a secondary form of identification, human face images help determine many other revealing attributes like gender, age, ethnicity and even the emotional state of a person. How does the effectiveness of multi-angled face segmentation effect face recognition accuracy, and how much training data is required to get an accurate and precise classification result?