3D Human Face Recognition as a Linearly Separated Problem

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


Neural Networks, Perceptron, Face Recognition, 3D Hu man Faces, Evolutionary Programming.


The objective of this article is to study patterns which represent human faces in 3D and are associated to lin ear separate class problems, where Artificial Neural Net work - ANN recognition approach is used. The first step of the study was an experiment which resulted in giving the conditions to the hypothesis of 3D human faces be long to the linear separate class. Low complexity nets that develop precise face classification task in "yes" and "no" classes were found through the use of evolutionary pro gramming in the architecture of three layers feedforward ANNs. Through this analysis, the use of Perceptron and Adaline proves, by tests, the linear separation on the human face recognition, because these ANNs are only capable to classify linearly separated patterns.

Important Links:

Go Back