Iris Location Based on Improved Vector Field Convolution

Abstract

The iris location is an important task for biometrics systems and which is critical for the iris recognition system. However, typical iris location methods always show the poor performance of huge searching range, noise sensitivity and computational cost. And we have discovered from research in recent years that not every iris image is a regular circle. In previous years, we proposed an precise iris location algorithm based on Vector Field Convolution (VFC) to improve the accuracy of iris location and solve the Non-circular fitting problem. In this work, the four directions Laplacian operator is proposed to settle the problem of location accuracy on the VFC model. The CASIA v1.0, MMU v1.0 and MMU v2.0 iris database are used to verify the accuracy of this improved algorithm. Compared with the traditional iris location algorithm and classical VFC method, this improved VFC snake shows its significant advantages, especially in improving location accuracy.