Design for an application specific processor to implement a filter bank algorithm for hand vascular pattern verification

This paper presents a design for an application specific processor (ASP) to implement a filter bank approach for a hand vascular pattern extraction algorithm, which is specifically designed for person verification applications. The implemented algorithm is based on a filter bank approach since it applies two different filters to the input hand images; one for effective extraction of normal-state blood vessels (normal mode filter: NMF) and the other for effective extraction of thin vessels and vessels in a contracted state (enhancement mode filter: EMF). We combine the outputs of both filters to obtain the final hand vascular patterns. The filter bank approach has been proven to prevent possible loss of vascular patterns due to variations in the external temperature. In addition, the proposed NMF and EMF are designed to have filter coefficients in the form of powers of two, which can be realized using fixed-point operators and which reduces hardware complexity dramatically. In order to validate the ASP design, we implemented a prototype system utilizing a field-programmable gate-array (FPGA) device. The prototype system showed shows approximately a five-times better verification error rate than the conventional single-filter algorithm. Also, the person verification speed, including feature extraction time, is measured to be 133 [ms/person].