An Example of Course Project

Face Identification

Agenda

Why Face Identification?

• Useful• Interesting• Creating own dataset for extra credits

Data set

• The Extended Yale Face Database B– 2414 images of 38 human

• Own data – 100 images of 4 people

Classifiers

• SVM (Support Vector Machine)– LIBSVM– Self Implemented SVM Optimizer

• ANN (Artificial Neural Network)

• Coded in Matlab

Classifier Parameters

• SVM – Kennel functions

• ANN – Layers and units

Feature Selection

• Raw pixels• Down-sampling pixels • Extracted features

Compare Kernel Functions

exp(|uv|^2)

u'*v tanh(u'*v)

(u'*v)^2 (u'*v)^360.00%

65.00%

70.00%

75.00%

80.00%

85.00%

90.00%

95.00%

100.00%

LIBSVM Multi-classifierImproved one-vs-all

Compare Feature Numbers

10 100 1000 10000 10000080.00%

85.00%

90.00%

95.00%

100.00%

105.00%

0

500

1000

1500

2000

2500

AccuracyTime

Feature Number

ANN Learn Rate

Report

• 10 pages

Schedule

3 2 2 3

Proposal

Prepare dataset

Data preprocessing

Algorithms implementation

2

Evaluation & Report

12 weeks total

Lessons Learned

• Start early• Review each other’s work