Super-resolution construction of IRIS images from a visual low resolution face video

Iris became an important biometric in the last decade, due to its uniqueness and richness of features. In this paper, a novel super-resolution and image registration technique for visual (non-infra-red) iris images is presented. In the proposed technique a full face, 3 second long, 90 frames, visual video is captured with a digital camera located 3 feet away from each subject. Iris images are segmented from the full face image. A cross correlation model is applied for the registration/alignment of full gray scale iris images. A high resolution iris image, that is 4 times higher in terms of size and resolution, is constructed from every 9 low resolution images. This process of building a high resolution image is based on an auto_regressive signature model between consecutive low resolution images in filling the sub pixels in the constructed high resolution image. Then this process is iterated until a 16 times higher resolution iris image is constructed. Illustrative images are shown that prove the effectiveness of the proposed technique.