Low power neuromorphic hardware based multi-modal authentication system

In this paper, we present a multimodal authentication system based on Face and Iris recognition, implemented on commercially available neuromorphic hardware. A novel variance based step for limbus segmentation is proposed in the pre-processing of iris data. The proposed learning/recognition technique uses a sequential bi-layered neural network with the initial stage relying on face match followed by sequential Iris match. The system is able to perform recognition (/per vector) in sub-50 μs with an accuracy of 97 %. Total on-chip recognition energy dissipation for both layers (/per input) was ∼ 1410 μJ.