Low-resolution Iris Recognition via Knowledge Transfer

This work introduces a novel approach for extremely low-resolution iris recognition based on deep knowledge transfer. This work starts by adapting the penalty margin loss to the iris recognition problem. This included novel analyses on the appropriate penalty margin for iris recognition. Additionally, this work presents analyses toward finding the optimal deeply learned representation dimension for the identity information embedded in the iris capture. Most importantly, this work proposes a training framework that aims at producing iris deep representations from extremely low-resolution that are similar to those of high resolution. This was realized by the controllable knowledge transfer of an iris recognition model trained for high-resolution images into a model that is specifically trained for extremely low-resolution irises. The presented approach leads to the reduction of the verification errors by more than 3 folds, in comparison to the traditionally trained model for low-resolution iris recognition.