Local Feature Based Face Recognition

Abstract

A reliable automatic face recognition (AFR) system is a need of time because in today's networked world, maintaining the security of private information or physical property is becoming increasingly important and difficult as well. Most of the time criminals have been taking the advantage of fundamental flaws in the conventional access control systems i.e. the systems operating on credit card, ATM etc. do not grant access by who we are, but by what we have”. The biometric based access control systems have a potential to overcome most of the deficiencies of conventional access control systems and has been gaining the importance in recent years. These systems can be designed with biometric traits such as fingerprint, face, iris, signature, hand geometry etc. But comparison of different biometric traits shows that face is very attractive biometric because of its non-intrusiveness and social acceptability. It provides automated methods of verifying or recognizing the identity of a living person based on its facial characteristics. In last decade, major advances occurred in face recognition, with many systems capable of achieving recognition rates greater than 90%. However real-world scenarios remain a challenge, because face acquisition process can undergo to a wide range of variations. Hence the AFR can be thought as a very complex object recognition problem, where the object to be recognized is the face. This problem becomes even more difficult because the search is done among objects belonging to the same class and very few images of each class are available to train the system. Moreover different problems arise when images are acquired under uncontrolled conditions such as illumination variations, pose changes, occlusion, person appearance at different ages, expression changes and face deformations. The numbers of approaches has been proposed by various researchers to deal with these problems but still reported results cannot suffice the need of the reliable AFR system in presence of all facial image variations. A recent survey paper (Abate et al., 2007) states that the sensibility of the AFR systems to illumination and pose variations are the main problems researchers have been facing up till.