Reconhecimento biométrico considerando a deformação não linear da íris humana

Abstract

The biometric systems that use the information on iris texture has received great attention in recent years. The extraordinary variation in iris texture allows the creation of recognition and identification systems with almost zero error rates. However, in general, researches ignore the problems associated with contraction and dilation iris movements that can result in significant differences between the enrollment images and the probe image.

This work, in addition to developing a traditional iris recognition system, comprising the steps of detection, segmentation, normalization, encoding and comparison, determines quantitatively the iris motion effect in recognition system accuracy. In addition, this paper proposes a new method to reduce the influence of dynamic iris, verified by decidability and the Equal Error Rate (EER), obtained in the comparison between iris codes in very different expansion states.

The new method uses Dynamic Time Warping technique to correct and compare the gradient vectors extracted from iris texture. Thus, the most discriminant features of the test image and enrollment image are aligned and compared, considering the non-linear distortion of the iris tissue.

Experimental results using dynamic images indicate that system performance gets worse with comparison on images in different states contraction. For direct comparison with contracted and dilated iris the proposed method improves the decidability of 3.50 to 4.39 and EER of 9.69% to 3.36%.