Handling pose variation in face recognition using SIFT

Abstract

This project deals with overcoming the limitations of the SIFT matching algorithm. We propose a novel way of taking the strengths of the SIFT algorithm and removing the underlying weaknesses to improve SIFT matching in pose variation. This is done by increasing the size of the search database to a certain degree of acceptance. In our bid to improve SIFT matching we have come up with an algorithm to find the pose of an image(albeit with some degree of error). We have stress tested our algorithm against many images and get reliably good results for both horizontal and vertical variation in pose. © 2011 IEEE.