A parallel hardware architecture for Scale Invariant Feature Transform (SIFT)

Abstract

Scale Invariant Feature Transform (SIFT) is an efficient algorithm for extracting distinctive features from images. It is used in many computer vision applications such as object recognition, motion estimation, robot mapping and navigation. Although it has an outstanding performance, its implementation requires extensive computations, and it is very difficult to achieve near real-time feature extraction using software implementation only. Hence, there is a clear advantage in exploring the feasibility of implementing the algorithm using customized hardware with the intent of achieving real-time performance. In this paper, a parallel hardware architecture is proposed to accelerate the SIFT features extraction. The proposed approach is viable and yields promising results in terms of accuracy, speed, and hardware resources.