Iris Super-Resolution for Images Sourced from Websites and Social Media

The iris is a highly distinctive and stable biometric trait widely used in high-security applications. Recent studies have explored iris recognition using ocular images collected from websites and social media. However, ocular images obtained from online sources exhibit significant non-idealities, one of the most critical being their lower resolution compared to images typically used for iris recognition. Super-resolution techniques can potentially enhance iris details and improve recognition performance. Nevertheless, existing studies on super-resolution for iris recognition focus on images acquired under controlled conditions and overlook images collected online. This work is the first to study super-resolution for iris recognition using web and social media images. We propose a transfer learning approach in which deep neural networks are trained on high-quality ocular datasets and then applied to ocular images collected online. Experimental results demonstrate improved recognition accuracy, especially for dark-colored eyes.