Super-Resolution Iris Image Restoration Based on Multiple MLPs and CLS Filter

Abstract

Iris recognition is a biometric technology which shows a very high level of recognition accuracy, but low resolution (LR) iris images cause the degradation of the recognition performance. Therefore, a zoom lens with a long focal length is used in an iris camera. However, a bulky and costly zoom lens whose focal length is longer than 150 mm is required for capturing the iris image at a distance, which can increase the size and cost of the system. In order to overcome this problem, we propose a new super-resolution method which restores a single LR iris image into a high resolution (HR) iris image. Our research is novel in the following three ways compared to previous works. First, in order to prevent the loss of the middle and high frequency components of the iris patterns in the original image, the LR iris image is up-sampled using multiple multi-layeredperceptrons (MLPs). Second, a point spread function (PSF) and a constrained least square (CLS) filter are used to remove sensor blurring in the up-sampled image. Third, the optimal parameters of the CLS filter and PSF in terms of the recognition accuracy are determined according to the zoom factor of the LR image. The experimental results show that the accuracy of iris recognition with the HR images restored by the proposed method is much enhanced compared to the three previous methods.