Reconhecimento de impressões digitais com aprendizado não supervisionado

• Autor(a) principal: Lopes, Gabriel Matthiesen [UNESP]
• Data de Publicação: 2017
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/156512; http://www.athena.biblioteca.unesp.br/exlibris/bd/capelo/2018-05-22/000898724.pdf
• topic: Computação gráfica; Biometria; Datiloscopia; Algoritmos

Due to the fact that fingerprints are unique, they are an effective way of identifying people. Currently, there are systems that use this way of identification, and its most common applications are in access control and criminal recognition. However, such systems, like any other, are prone to errors, and considering their significant influence in matters related to security, it is important to reduce its occurences. Therefore, there is a great interest in ways to produce more correct results, which motivates development of more effective algorithms as well as other methods that allow errors to be filtered. In this paper, a CBIR(Content-Based Image Retrieval) approach is used in fingerprint images provided by the Fingerprint Verification Competition (FVC) editions of the years 2002 and 2004. This approach provides ordered lists, in which the images most similar to the query image appear in the first positions. Then, on this list, an unsupervised learning method, called RL-Sim*, is apllied. The results, with and without the use of this method, were compared to evaluate the effects of this post-processing phase on the final results. This comparison evidenced that applying RL-Sim* caused a general improvement on the results, demonstrating that it is an interesting option to fingerprint identification