Scale Invariant Feature Transform algorithm based on a reconfigurable architecture system

This paper proposes a reconfigurable solution to accelerate the Scale Invariant Feature Transform (SIFT) algorithm, since the reconfigurable computing possesses the advantaages of both high performance and flexibility by combining the characters of programmability of GPP and specific of ASIC. We also present several techniques, such as polyhedral model, block technology etc, to improve the execution performance on a REconfigurable MUltimedia System called REMUS. Tests show that the execution performance of SIFT is improved by 6% and 60% comparing with that executed in the multi-core platform and FPGA separatively with high flexibility and low power consumption, when utilizing a 200MHz working frequency which is suitable for embedded system.