Applying Dimensionality Reduction for Neural Networks Learning in the SORFACE Project

L.S. Encinas, A.C. Zimmermann, J.M. Barreto, and L.O. Marin (Brazil)


3D Faces, Neural Networks, Dimensionality Reduction,SVD, PCA


Face recognition is an easy task for humans, however computers cannot solve this problem quickly and easily be cause they analyze all the information from the face im age. This article presents a study case, inserted in the Optic System for Human Face Recognition SORFACE project, where dimensionality reduction techniques are used over the face set, in order to reach a better neural network learn ing time, gathering the necessary information to be able to the face identification. Two tools are used to lower the com plete face set dimension. The Principal Component Anal ysis PCA, where each component carries a variance por tion from the face set covariance matrix. And the Singular Value Decomposition SVD, the third great matrix factor ization and one of the best ways to compute eigenvalues and eigenvectors. These techniques assist the recognizing system, improving the learning time since the number of inputs in the neural network is directly proportional with the dimensionality of the face set.

Important Links:

Go Back