FACE RECOGNITION USING FILTERED EOH-SIFT

This paper presents a method for the implementation of facial recognition using filtering techniques that will increase the accuracy of the process as well as distinguish faces more decisively. The process has proved to be invariant to image scale, rotation and illumination. This paper was motivated by the EOH-SIFT approach. The process described in this paper is aimed at improving upon the effectiveness of EOH-SIFT by feeding it a filtered image. After much exhaustive research into various filters, this paper shows that, with two filters which when used in a specific order, significantly boost the potency of the EOH-SIFT approach to identify faces. This approach has given very promising results when tested on the ORL database. Although it is a standard dataset, it does not have much variation as seen in the real world. Hence, our approach has been tested on other datasets and it has produced extremely encouraging results. The recognition of faces proceeds by first obtaining the region of interest and applying the filters on that area followed by the identification of important features in the faces and then matching them using an efficient nearest-neighbor algorithm. This leads to a robust and definitive face recognition system. (C) 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Organizing Committee of ICCCV 2016