.
Support vector machines for face recognition? Abhijeet Jivendra Jha
Support vector machines (SVMs) are set of
supervised learning methods that are mainly used these days popularly used for
classification and regression. SVMs view input data as two sets of vectors and
then construct ahyperplane in that space that provides a margin between the
two. An optimal margin is calculated by first constructing two parallel
hyperplanes each to the side of the separating hyperplane and then are pushed
up against the data set. Since the larger the margin, the lower is the
generalization error of the classifier, the hyperplane with the largest
distance to the neighboring data points of both classes provides a good
separation . SVMs are quickly becoming very popular in the machine learning
community as a technique for tackling high-dimensional problems.
Face
recognition as a learning problem has received a lot of attention lately. One popular
method involves reducing the dimensionality of the problem using PCA and then
selecting the nearest class (eigenfaces).
http://www.cse.unr.edu/~bebis/MathMethods/PCA/case_study_pca1.pdf
Fisherfaces and JPRC are two
other popular alogorthms for face- recognition. More on these can be found at:
http://www.cs.cmu.edu/~rahuls/pub/fg2000-rahuls.pdf
http://www.cse.unr.edu/~bebis/MathMethods/LDA/belhumeur.pdf
Project
Motivation:
SVMs are yet to be applied to the problem of face recognition. The ultimate goal of the project would be to implement an SVM for this problem. In this process issues like how best to apply SVM to the n-class face-recognition problem, what would be the best strategies for training and preprocessing of the images and finally, comparing how SVMs perform to other techniques would have to be addressed. This will involve acquiring performance data for the eigenfaces, fisherfaces and JPRC techniques. An implementation of SVMs is available as Thorsten's SVMlight. http://download.joachims.org/svm_light/c...indows.zip
ORL and FERET are popular databases for
face-recognition. Few of the many other available databases are:
The Color FERET
Database, USA
The
Yale Face Database
PIE
Database, CMU
Project
- Face In Action (FIA) Face Video Database, AMP, CMU
The
Extended M2VTS Database, University of Surrey, UK
Proposed TimeLine:
Rest of April Review literature, theory,
state of Art
By May 12 Figure out and implement
training/pre-processing strategies
By end of May Have some experimental
results/ comparisons ready
June 2 Finish write up/presentation