Face recognition using eigen faces and direct cosine transform approach

Abstract

Face is a complex, multi-dimensional and meaningful visual stimuli for which it has been extremely difficult to construct a robust model for face recognition. In this thesis, Face Recognition is done by Eigen Faces Approach and by Discrete Cosine Transform (DCT) approach. Face images are projected onto a featured space called ‘Face Space’ that encodes best variation among known face images. The Face Space is defined by Eigen Face which are Eigen Vectors of the set of faces in the database. The DCT approach exploits the feature extraction capability of the Discrete Cosine Transform invoking both geometric and illumination normalization techniques which increase its robustness to variations in Facial images such as variation in scale, orientation, illumination variation and presence of some details such as dark glasses, beards and moustache etc. These methods were tested on a variety of Face Databases such as The Achermann Database, The Olivetti Database and The MIT Database having different variations and the results were also observed.