Abstract

Comparison of Feature Extraction Techniques for Face Recognition

Author(s): Davoud Aflakian1*, M. Syamala Devi2andEkta Walia3

Face Recognition (FR) is the use of computer to automatically verify/identify faces irrespective of various aspects such as pose, rotation, expression, scale and illumination. It is not possible to perform FR irrespec-tive of above aspects using any single algorithm. The paper presents a comparative study of different algorithms for feature extraction that is very important component in FR. The feature extraction algorithms reduce image dimen-sionality by extracting significant features from large number of pixels of the face, resulting in reduced time com-plexity and increasedaccuracy. There are number of local, global (holistic) and geometrical model based approaches for feature extraction in time and frequency domains. The focus of comparative study is based on local and global features in the time domain as well as frequency domain. The advantagesand limitationsof various feature extrac-tion algorithms for FR are highlighted. The suitability of using an algorithm for specific above aspects is also men-tioned. The comparison is performed based on review of literature as well as implementing algorithms and testing on ORL database. Euclidean distance is applied on extracted features forclassification. It is concluded that frequen-cy domain based techniques are more robustto variation in pose and expression as compared to time domain. Also, the Phase components are robust to illumination.

PDF