High performance Iris Recognition System on GPU

Iris Recognition stands out as one of the most accurate biometric methods in use today. However, the iris recognition algorithms are currently implemented on general purpose sequential processing systems, such as generic central processing units (CPUs). In this work, we presented a more direct and parallel processing alternative using the graphics processing unit (GPU), which originally was used exclusively for visualization purposes, and has evolved into an extremely powerful coprocessor, offering an opportunity to increase speed and potentially enhance the resulting system performance. Within the means of this system, the most time-consuming stages of a modern iris recognition algorithm are deconstructed and directly parallelized. In particular, template matching and identification are parallelized on a GPU-based system, with a demonstrated speedup of 15.6 and 10.7 times, respectively, and 1.3 when taking into account all system stages, compared to that of CPU-based version. We specifically implemented an Iris Recognition System based on Daugman's System for training and classification in C#. We executed the CUDA-C code on a NVIDIA GTX 460 Fermi 336 cores card. Our implementation of iris recognition could simultaneously estimate values for 2K test patterns in about 11 ms based on an input data set of 20 M patterns.