Celebrity Face Matching Using Deep Learning

Face detection and face recognition are hot topics in computer vision and have many real-life applications. In this paper we present a celebrity face matching system to match a random human face with celebrities' faces taken from the Pins Face Recognition dataset. Several deep learning models for face detection and face recognition are explored and compared. An initial accuracy of over 89.4% is achieved. A simple face alignment method is then applied which increases the accuracy to over $\boldsymbol{99.26}\%$, We also apply K-Nearest Neighbours with cosine distance and 100 neighbours to make more robust predictions.

I. Introduction

Deep learning has become more and more prevalent in recent years. Face recognition is one key area of application. In face recognition, computers are given images or videos as inputs with the goal to identify people's identities by comparing their faces with labelled faces in a database. Due to the emergence of powerful image recognition techniques based on deep learning and the increasing availability of facial data, face recognition techniques have evolved rapidly, e.g., for identity verification used by mobile banking etc [12–16].