Human Face Reconstruction and Recongnition

A Linear Algebra Approach

Semester Project by Kevin Ng and Suresh Subbiah
for CS395T: Large-Scale Data Mining

Project Overview

In this project we have analyzed various schemes for automatic human face reconstruction and recognition.
Both these problems have been well studied in literature but recently reserch on these topics have taken a
new path. Following the 1990 paper by Sirovich and Kirby, linear algebraic techniques have been used to
reduce the complexity of the problem. In this report, following Professor Dhillon's suggestion, we have
introduced two computationally cheap methods to tackle the problems of face reconstruction and recognition.
We call them Meanfaces and Svdfaces. We also compare the results obtained by these methods with
results obtained by applying Eigenfaces and Fisherfaces methods to the same dataset. Eigenfaces and
Fisherfaces are classical methods in this field now. References to these approaches may be found
in the class reading list.

Project Reports

Proposal : Face Recognition, February 5th, 2000
Mid-term Report : Human face Reconstruction using Eigenfaces, March 22nd, 2000
Final Report : Human face Reconstruction and Recognition, May 8th, 2000

Software and Face databases

During the course of this project we have written software which can be on MATLAB. There are three
major code segments. The first two are GUI based and allow the user to see the faces being analyzed.
The last code segment is suitable for collecting data over large data sets, by considering diffrent
partitions of the dataset into training and test faces. Each code segment comes with a README file.

Eigengaces GUI

Fisherfaces, Meanfaces, SVDfaces GUI

Datacollecting code

We also provide links to the two face databases we used for the project

MIT database (143 persons, 1 image each, 128 * 128 pixels, bmp format)

Yale database (15 persons, 11 images each, 121 * 160 pixels, bmp format)