Recognition of Human Faces Using Scale Invariant Feature Transform

Abstract

Automated facial recognition is a computer application for automatically identifying or verifying a person from a digital image or a video frame from a video source which is becoming an increasingly important for many areas such as human machine interfaces, multimedia, communication, visually mediated interaction, anthropomorphic environment and especially in security purposes. Humans have always had the innate ability to rec- ognize and distinguish between faces, yet computers only recently have shown the same ability. In the mid 1960s, scientists began work on using the computer to recognize human faces. Since then, facial recognition software has come a long way. There were so many face recognition researches have been done regarding still image as well as video based. This research presents a new approach for face recognition based on Lowes Scale In- variant Feature Transform algorithm. SIFT feature, by David Lowe, is a robust invariant local feature that have the property of scale and rotation invariance. Also, it is able to pro- vide accurate matching over a certain range of a ne transform and illumination changes. Thus, it could be useful to extract distinctive face features over di erent scales, rotations, views or expressions of human faces for the detection task. In SIFT approach four major stages of computation should follow to generate the set of image features. The rst stage of computation is Scale-space extrema detection. Second stage is Key point localization and Third stage is Orientation assignment. Final stage is Key point descriptor. The local image gradients are measured at the selected scale in the region around each key point. These are transformed into a representation that allows for signi cant levels of local shape distortion and change in illumination. In this research First SIFT feature are extracted for a set of given image which is implemented using MATLAB. Then these are stored in a separate le and names of the le are stored in database with details of the relevant person. For a given original image SIFT features are extracted rst and compare them with the features of the database. Then nd the best match based on the nearest neighbor approach and taken the personal details of the matched person. To practically test, application is implemented using mobile phone and pc. For a given frontal view image and di erently scaled image system performs 100% accuracy. Also for the images with slightly rotated, system gives 100% correct results. But for mostly rotated face images, it gives only 40% of accuracy level. This shows that SIFT is partially invariant to rotation. For a given image with noise, system performs accuracy level of 90%. So considering the overall performance, system gives average of 86% accuracy in recognizing the faces. This system is very useful as it costs less and gives highest accuracy results.