Multimodal Biometrics

2D Face Recognition

This project aims to develop a security platform capable of detecting and recognizing people through 2D face images. The system must use images from low cost video cameras, that makes possible its application in different environments. It also will be integrated in the GDM for users authentication under Linux.

The segmentation uses as classifier sets of Haar Features, selected from trainings. The recognition is made with eigenfaces, calculated with Principal Component Analysis (PCA).

The system is in phase of development and tests.

Face Detection
Face Recognition

3D Face Recognition

The main goal of this project is the development of a framework for face authentication from 3D images. The main goal is automatically verify if a subject is who he claims to be. The framework is divided in five stages: (1) 3D Face image aquisition; (2) Image pre-processing; (3) Feature extraction; (4) Registration; (5) Evaluation of of the error measure.

The facial images are aquired using a laser scanner (1). On stages (2) and (3), the region of interest is extracted from the facial image automatically. The images are registrated using Simulated Annealing (SA), on stage (4). On the next stage it is applyed the Surface Interpenetration Measure (SIM) to evaluate the facial recognition. Provided two 3D face images, we can verify if they belong to the same subjects with a verification rate of 99%, at a False Acceptance Rate (FAR) of 0%. At this FAR, the possibility of identifying a non-authorized individual in an authentication system is eliminated.

3D Face Image
3D Face Segmentation


The objective of this project is to develop a new approach for fingerprint matching based on image alignment. This new approach uses Simulated Annealing (SA) and Iterative Closest Point (ICP) to alignment the input and template fingerprints using the ridges structure. The main objective is to provide a fast and accurate identification system. This system is capable to use small images, and consequently, to use a reduced number os minucias, to perform the recognition based on these features. The results show that the acceptance rate and the performance of our biometric system are as good as the presented results of the state-of-art approaches based on ridges.

This biometric system has two important modules: pre-processing and authentication. In the pre-processing module, features are extracted of the fingerprint to generate a biometric model associated to a specific subject. In the authentication, the system search in the database the template associated to this subject and compute the matching score to verify the identity.