Iris matching algorithm on many-core platforms

Abstract

Biometrics matching has been widely adopted as a secure way for identification and verification purpose. However, the computation demand associated with running this algorithm on a big data set poses great challenge on the underlying hardware platform. Even though modern processors are equipped with more cores and memory capacity, the software algorithm still requires careful design in order to utilize the hardware resource effectively. This research addresses this issue by investigating the biometric application on many-core platforms. Biometrics algorithm, specifically Daugman's iris matching algorithm, is used to benchmark and compare the performance of several manycore platforms. The results show the ability of the iris matching application to efficiently scale and fully exploit the capabilities offered by many-core platforms and provide insights in how to migrate the biometrics computation onto high-performance many-core architectures.

An FPGA-based Hardware Accelerator for Iris Segmentation

“…[4,26,32] are references to works that use wavelet transforms. A simpler alternative to the wavelet transformation approach is the Ridge-Energy Direction (RED) method originally presented in [19] and further used in [24,28].…”

GPU Processing for Biometric Big Data Based Identi.ication. Why and what for?

“…In [26] the GPU is applied only to the iris segmentation phase. Additional approaches to iris-based matching have been presented by [27] where the energy consumed is also taken into account and [28] where very modest GPU hardware outperforms a multithreaded CPU based solution.…”

Rocket: Efficient and Scalable All-Pairs Computations on Heterogeneous Platforms

“…All-pairs compute problems, which evaluate a function for each pair of items of a data set, are prevalent in many scientific domains including radio astronomy [1], microscopy [2], bioinformatics [3], digital forensics [4], computer vision [5], data mining [6], information retrieval [7], and biometrics [8]. These problems typically involve calculating some measure, such as the distance or similarity, between pairs of data items, such as images or objects.…”