Sistemas de identificação pessoal utilizando tecnicas de reconhecimento e verificação facial automaticas

AUTOR(ES)
DATA DE PUBLICAÇÃO

1997

RESUMO

This thesis dissertation shows the development of biometrics systems based in face recognition technics. These biometric systems are able to recognize and identify people by extracting information from their facial images. Face recognition systems must have a model to encode and represent facial image information for recognition purposes. We choose the eigenface method [fURKJ, due to its remarkable capability to extract and compress facial image information, as the core for the feature extraction task. The eigenface method was implemented in a semi-automatic way. The eigenface system framework is tested in two distinct tasks: face recognition and face verification. To perform the tests, we acquired a face image database composed of 435 views of 102 people. No hard constraints were imposed to the database acquisition process what means we obtained large variations between images of same people. We also evaluate the system s tolerance against such variations. Ali the routines needed to the task of image processing, face recognition and face verification were written in C++. They compose a library of routines that can be adapted for other task such as automatic segmentation of facial images. The face recognition performance is evaluated against the fuU database, 435 images, trying to classify each image in its correspondent class. Each class represents one person present in the database. The face recognition system achieved a recognition rate of 97.70%. The face verification system was tested with the fuU database, 435 images, using a minimum distance classifier varying its acceptance threshold

ASSUNTO(S)

analise de componentes principais reconhecimento de padrões processamento de imagens

ACESSO AO ARTIGO

Documentos Relacionados