High dimensional reduction algorithm based facial recognition system

Abstract

This project presents a high dimensional reduction algorithm based facial recognition system, in both desktop and cloud version. In this project, Principal Component Analysis (PCA) is employed as the high dimensional reduction algorithm. Face images are represented by selected eigenfaces and weight vectors, which can largely reduce the dimensionality of face set. Although it involves a lot of matrix calculations, the systems are developed mainly in Java platform instead of technical computing platform like MATLAB, so they can be used for commercial purposes. In desktop version application, users can easily choose their cropped face set, train it and then conduct facial recognition by probe images. However, this version requires users to pre-process the images before they do facial recognition, which limited the use cases it can be applied to. To further improve the system, and make it capable in different user scenarios, a cloud version system is developed. The server side application in Apache Tomcat follows Representational State Transfer (RESTful) style, and mobile client is in Android. The dimensionality reduction performed in server side, then the trained set will be cached. When client side send recognition operation with facial image, the server side will perform recognition using 1-nearest neighbour algorithm, and then return the result to client side.