Open-Set Face Classification for Access Monitoring Using Spatially-Organized Random Projections

Abstract

In this paper, we present an easy method of open-set face classification problem with application to access control and an identity verification. We use normal random projections as a method of feature extraction from face images. The image transformation consists of local projections of spatially-organized rectangular blocks of an image. Two classification algorithms are analyzed: the nearest neighbor method with scalar product similarity measure and individual acceptance/rejection thresholds and multinomial logistic regression. The computational complexity of designing the transformation is linear with respect to the size of images and does not depend on the form of image partition. Experiments performed on the ORL Database demonstrate that the proposed technique is simple and suitable not only for an access monitoring system but also for face verification. The contents of an image after RP-based transformation is hidden and will not be stably recoverable. So, this approach can be used in systems where the privacy-preserving property is important.