Enhancing Face Recognition through Overlaying Training Images

Abstract

In the usual face recognition approach, system is getting trained through a large number of training samples. It means in the process of training, features are extracted from all the training images individually. In this process many redundant features are required to be eliminate also. During feature elimination, some features also get suppressed due to inappropriate thresholds. So, this approach is typically time consuming and costly in the part of training. Hence, there is a requirement of feature extraction in such a way that it reduces the chance of data redundancy and system complexity. This paper presents a facial recognition technique by inclusion of superimposed version of all relevant images which improves the accuracy of the model by roughly 43 percent. The algorithm aims to establish the importance of superimposition strategy in the field of face recognition. The Haar feature based classifier is used, where a cascade function is trained from a set of images. We have used the open source database of faces from the archives of AT&T Laboratories Cambridge to train and test our model.