An Investigation into the Effects of Image Resolution on a Facial-Image-Based Personal Authentication System

Abstract

The issues associated with image resolution in automated authentication or identification systems has become one of the important challenges for researchers' in biometrics. The aim of this thesis is to investigate the effect of variable resolutions on the performance of a Facial-Image-Based Person Authentication System. Image resolution may vary significantly especially in uncontrolled acquisition environments or when sensing from a distance and so on. The detail available in the data thus reduces which may deteriorate the performance of such system. In this project we investigated the impact on system accuracy when image resolution is gradually reduced by a given factor. As a remedy, we investigated different methods for increasing image resolution prior to using those images for authentication and compared the relative gains in accuracy. The main procedure of the face image authentication system based on comparing landmarks of the face remains the same.

In this study, we found that several issues related to image resolutions might have an impact on the recognition rate performance such as facial expressions, image background, and others. The influence of image resolution on the recognition rate increases roughly with the increasing resolution at a specific degree, high-image resolution would not be good for recognition rate always; reducing high image resolution makes it easier to achieve high face recognition rates.