Reconhecimento biométrico de íris usando filtro de correlação

Abstract

Naturally, patterns that we wish to recognize occurs in several manners. As example, considering the ordinary human iris, it is often to this pattern to manifest different aspects. When acquired one iris can present, in relation to the original pattern, rotation, translation, lighting distortions or mixed noise. Thus, in some cases, it is necessary to the recognition method to have versatility enough to identify such unexpected forms of pattern occurrences that we wish to recognize. In our particular case, filters for correlation with broader flexibility than the standard SDF and MACE filters, for example, are necessary. A consolidated way to express the variation of a data set can be obtained using the method of Principal Component Analysis (PCA) [1, 2]. The PCA optimally represents a dataset and this fact makes it interesting for combination among ordinary correlation filter design and one possible approach consists of the modification of the design of correlation filters to use principal components as their own detection target. In this dissertation, it is proposed to modify the design of correlation filters SDF and MACE to use Principal Component Analysis to represent the set of occurrences of the pattern of interest. One benefit of this approach resides in the fact that PCA incorporate the changes in the set data providing as a result, more flexible filters. Thus, filters designed in this way would succeed in detecting patterns with small distortions translations and rotations. To validate the proposed method, a recognition system that uses the default human iris was designed and, for this purpose, one consolidate database is used.