A Comparative Analysis of Massive Finger-Vein Recognition Algorithms: From Energy Consumption Perspective

The volume of data generated has grown exponentially in recent years worldwide. In general, parallel architectures such as clusters, multicores, and GPUs have reduced processing times compared to their sequential version. However, in addition to the computational efficiency expressed in processing time, the scalability of the systems, the energy efficiency, and the price of the electrical energy consumed must be considered in parallel systems. This work aims to present an analysis of energy consumption between a GPU and a multicore platform, using the same exhaustive search algorithm as a case study to solve kNN queries in metric spaces on a database of finger vein images. The experimentation was performed on a database of 4,000,000 finger vein images on multicore and GPU platforms. The experimental results show that the GPU platform is 2.52 times lower than the multicore platform in dissipated energy when solving 56 simultaneous queries. The conducted study is pioneering in this kind of analysis on massive human recognition tasks.