3D FACE RECOGNITION

M N PRATYUSHA RAO 07D21A1228 IT 3rd YEAR

C. SRUTHI 07D21A1249 IT 3rd YEAR

Abstract

an initiative for future advances. Our paper is a description of the face recognition technology. Biometric face recognition is currently used in handful of applications to guard against the perceived threat of hackers. A number of groups are looking at using this technology in the near future.

Biometrics is a generic term that refers to a wide range of measures of biological data. The use of biometrics is not new. The signature has been used for authentication for a very long time. Later the system was fine tuned to a photograph. A whole range of well-known and new biometrics is being used and experimented with.

Biometrics is the process of identifying an individual using his physical feature(s) or action(s) which are both unique and measurable. Real life applications like ATM, AIRPORTS, and BANKS etc lack proper security as the logical elements of security like passwords, ID cards are easily captured. The ambiguity in using them can be relieved using various biometric technologies.

Introduction: Biometric face recognition works by using a computer to analyze an individuals facial structure. Face recognition software takes a number of points and measurements, including the distances between key characteristics such as eyes, nose and mouth, angles of key features such as the jaw and forehead, and lengths of various portions of the face. Using all of this information, the program creates a unique template incorporating all of the numerical data. This template may then be compared to enormous databases of facial images to identify the individual.

Latest advances in the field of biometrics can be categorized into technologies like face recognition, finger scanning, retina patterns, hand geometry,

Voice recognition etc. Combination of these technologies with the logical identities ensures a remarkable security to the applications.

Understanding the technologies not only ensures proper utility but also provides

FEATURES OF FACE RECOGNITION Primary advantage of using this biometric system is its ability to operate hands-free and users identity is confirmed by simply staring at the screen. Its completely a software based system. Uses two methods for capturingvideo or thermal imaging. Continuous monitoring of the user. Access to sensitive information can be disabled when the user moves out of the cameras field. Verification is then performed when the user returns to work at the desktop. PROCEDURE Face recognition essentially operates in four-stage process that is automated and computerized. It can be seen in the figure represented. CAPTURE A physical or behavioral sample is captured by the system during enrolment.

The template is then compared with a new sample. The biometric data are then stored as the biometric template for that person.

MATCH/NON-MATCH The system then decides whether the features extracted from the new sample are a match or a non-match with the template. When identity needs checking, the person interacts with the biometric system for a second time, a new biometric sample is taken and compared with the template. If the template and the new sample match, the persons identity is confirmed.

EXTRACTION Unique data are extracted from the sample and a template is created. Unique features are then extracted by the system and converted into a mathematical code. This sample is then stored as the biometric template for that person.

COMPARISON

1. State of the art of 3D face recognition Till recently, most research in face recognition had focused on 2D intensity images. The primary reason for this bias towards 2D face recognition was because they were easy to acquire. Further, quite decent results had been achieved by using 2D face images in constrained environments where illumination was assumed to be constant. However, a number of recent studies have shown that the performance of various 2D face recognition techniques is adversely affected by varying lighting conditions. 3D face recognition is invariant to such environment changes (assuming 3D reconstruction algorithms can handle variance in environmental conditions) and thus has become the epicenter of biometric face recognition system research. Most 3D face recognition systems generate 3D models by multiple calibrated cameras. This approach is both slow and infirm. We propose to use Canastas Electronic Perception Technology for building a Biometric face recognition system. The time of flight based mechanism to acquire the 3D face is best suited for such an application. 3D face recognition research is still weakly reported in the published literature. This is primarily because of the fact that 3D capture based on reconstruction from multiple 2D images requires sophisticated hardware such as calibrated and synchronized cameras. This makes recovery of 3D data, an extremely slow and expensive process. Some researchers have also used other data extraction methods like the 3D laser scanner, which though being highly accurate are still extremely slow.

2. An Overview of Commercial 3D Face Recognition Systems

A few commercial 3D face recognition systems have been unveiled recently. The most important among them being by NEC and MERL. We give an overview of the pros and cons of these systems. The MERL system uses 3D face models and specifically targets illumination and pose variation, which they think as the most critical factors limiting performance. They capture the 3D shape of a human face from a sequence of sparse 2D silhouettes from multiple cameras (or video) with no manual user interaction. Their silhouettes based

approach decouples the geometric subtleties of the human face from the nuances of shading and texture. Their system is able to obtain decent 3D face models from silhouette intersection (which is being increasingly used in computer vision research these days). However, the tests conducted by them till now have been in favorable environments, and it is not clear how their approach would be able to cope with nosy data. NEC claims that its system is the world's most accurate personal identification system. They use GIB (Geodesic Illumination Basis -descriptors describe differences in illumination on the skin) descriptors as registered data, which is calculated from 3D facial data, enabling optimal description of various pose and illumination changes. The key features of their approach is the matching method which identifies and highlights individual facial characteristics of 2D images and compares them with 3D and GIB facial data. They claim to have achieved accurate personal identification matching rate of 96.5%, even under very severe environmental conditions that cause changes in illumination and pose. They have conducted tests using a database with registered 3D facial images of 1000 people.

Fig. The average face and first five eigenfaces computed with no image pre-processing

3. The 3DFACE system In the Geometric Image Processing laboratory (Department of Computer Science, Technion) designed a prototype 3D face recognition system based on the expressioninvariant representation of facial surfaces. Current 3DFACE system prototype is shown in Fig.

It

operates both in one-to-one and one-to-many modes. The 3DFACE system runs on a dual AMD Opteron64 workstation under Microsoft Windows XP. One of the CPUs is dedicated merely to processing and computation of the canonical forms; another one handles the graphical user interface (GUI) and the visualization. Data processing in the 3DFACE system can be divided into several stages. First, the subject's face undergoes a 3D scan, producing a cloud of points representing the facial surface. The surface is then cropped, smoothed and sub sampled. Next, a feature detector is applied in order to find a few fiducial

points. Next, a geodesic mask is computed around these points. Finally, the facial surface undergoes Canonization using an MDS procedure.

Smoothing

(a) Before smoothing Fig. Scheme of the 3DFACE system pipeline.

(b) After smoothing

Feature detection

The processing stages in 3DFACE system pipeline are 3D acquisition AND cropping

Geodesic mask

challenge in the automatic facial-expression analysis domain. We will need to incorporate a dynamic model in our framework for 3D face recognition to realize such a system. There are a couple of benefits of adding a dynamic model to our framework. First, we will be able to track alignment changes in our input image, making out matching task easier. Second, we will be able to refine our 3D model as we move over multiple 3D faces of low resolution. 5. Conclusions presented a geometric framework for three-dimensional face recognition that naturally incorporates the nonrigidness of the facial surfaces and allows coping with facial expressions. Our isometric model reduces the problem of comparing faces in the presence of facial expressions to the problem of isometric surface matching. We use the isometric embedding as a method to construct isometric invariant representation of facial surfaces, which gives an expressioninvariant representation of the face. An implementation of our 3DFACE algorithm shows high recognition accuracy, significantly outperforming standard 2D and 3D approaches, and working favorably even in the presence of very strong facial expressions. The main advantage of the 3DFACE algorithm is definitely when a large variability of facial expressions is present. The significance of it is first of all in commercial applications, which can never assume collaborative users, and especially in cases when the face recognition is assumed to work in a natural environment. Ideally, the user will be unaware of being scanned and recognized, which implies that the variability of his facial expressions can be significant. We

Fig. Geodesic mask computation and examples of the geodesic mask insensitivity to facial expressions

Canonization and surface matching 4. Possible Extension to Facial Gesture Recognition

Automatic analysis of facial gestures is an area of intense interest in the human-computer interaction design community. Facial gesture capture data can be used to drive animation sequences. A robust way to discern facial gestures in images of faces, insensitive to scale, pose, and occlusion, is still the key research

Besides being insensitive to expressions, canonical forms conceal some other favorable properties. First, the obtained representation is irreversible. Therefore, given the canonical form it is impossible (or at least very hard) to find the underlying original facial surface. Thus, the canonical form in some sense hides the actual identity of the subject stored in the gallery. This is significant in commercial systems where the security of the biometric data is an important issue. Secondly, canonical forms provide an intrinsic parameterizations of the facial surface, which leads to an easy registration of the facial images, and consequently, to an easy fusion of 2D and 3D information. Thirdly, embedding has a regularization effect on the facial surface: small local artifacts lead to fluctuation of all geodesic distances. The fluctuations in the canonical forms are no more local, but rather spread among all points of the canonical form. As a consequence, the canonical forms are less sensitive for example to acquisition and processing artifacts than the original surfaces. 6. References 1. Achermann, B. and H. Bunke: `Classifying range images of human faces with Hausdorff distance'.

2. Besl, P. J. and N. D. McKay: 1992, `A method for registration of 3D shapes'. IEEE

3. Introduction to biometrics 4. Some websites related to biometrics