Efficient heterogeneous face recognition using Scale Invariant Feature Transform

Abstract

Face recognition includes analysis of an image and extracting its facial features which will help to discriminate it from others. Scale invariant feature transform (SIFT) to extract distinctive invariant features from images can be used to perform reliable matching. The features extracted are invariant to rotation, image scale and illumination. Systematic investigation of face recognition using SIFT features has been done. Being high distinctive features, every feature can be matched correctly with high probability against a huge database of features from many images. Result shows that SIFT is flexible recognition algorithm as compared to Contour matching algorithm for heterogeneous images. Both the algorithms are experimentally evaluated on AT&T, YALE and IIT-KANPUR databases with moderate subject size. Though Contour matching provides computational simplicity, SIFT provides efficient face recognition technique under pose, expression and varying illumination condition. Experimentally it confirms that Contour matching outperforms when the database is small but for large databases Scale Invariant Feature Transform (SIFT) gives more than 90% recognition rate.