A Component Based Face Recognition Technique using One Shot Learning for Forensic Application

Abstract

Face recognition is a technique of verifying or identifying the identity of a person by their face and it is one of the significant issues in object recognition and computer vision. It is an important and regular forensic tool used by criminal investigators. An essential part of human beings is the face and requires detection and recognition for various applications such as security and forensic investigation. However, due to the difficulties of different facial expressions, position variations, occlusion, aging, and resolution in image or video sequencing images, it requires proper techniques for face detection and recognition. Hence, the systems working on components of the face or facial landmarks have gained great importance. Due to the variety of criminal activity, accurate and effective identification has become a basic requirement for forensic application. Forensic examiners differentiate different facial areas of face images acquired from uncontrolled and controlled conditions captured from the suspect. This research focuses on Facial-landmark based Forensic Face Recognition. This study proposes a method by using Component-based Face Recognition with the information acquired from component of faces, by combining MobileNet-v2 with a Siamese Neural Network to increase the accuracy of face recognition rate with faces of different pose variation, expressions, and illuminations. The experiment results show that our proposed framework of face recognition by different facial expressions, poses variations, and illumination image data on the GTAV database of over 44 persons, this study have achieved a recognition accuracy rate of 95.4%. The Self-prepared face database obtained a recognition accuracy rate of 93%. This Paper gets better performance compared to face recognition without components, One-shot with Autoencoder, and One-shot without Autoencoder.