A General Approach on Facial Feature Extraction and Face Attributes

Facial feature extraction is method to recognise the nose, mouth, eyes and other components present in the human face. These components are very important in face recognition, face matching, expression recognition, classification and face tracking. Facial recognition system detects and returns the information regarding visual content with confidence scores in human face from digital image or frames from video source. To determine the series of human face attributes various methodologies and algorithms are proposed and developed. Facial Recognition results include both visual biometrics and wearables around the face like headphones, eye glass, hat and mask. The system accepts image as input and processes it with either cloud based dataset or local capabilities. It then extracts more than twenty seven facial information which includes gender, mood, visual age, smile expression, face orientation, color, picture quality and many more. This metadata can be used in many realtime applications such as biometric security, mobile camera technology for face filters and animoji, photo database services and social media profile moderations with content analysis. Hence all these methods are accessible by automated systems. Digital image will be processed to identify the facial landmarks along with information presents and returns the accurate attributes scores. This paper present brief study on facial feature extraction methods and attributes. Mentioned attributes are obtained from Microsoft FaceAPI, FacePlusPlus Cognitive Service Face++API and AmazonWebService Rekognition.

I. Introduction

Soft biometrics is a popular emerged automated and reliable technology to recognise, identify and classify Human behaviours [1]. A significant growth is observed in computer vision technology [2], where object detection has become a popular research area, that has a wide range of applications in day to day life. Facial recognition is a deep learning face print technology for verification and biological identities [3]. Where 1 to 1 face matching used in verification and 1 to many matching used for identification.