Asynchronous Processing for Latent Fingerprint Identification on Heterogeneous CPU-GPU Systems
MetadataShow full item record
Institute of Electrical and Electronics Engineers (IEEE)
Asynchronous processingAccelerator architecturesCUDAFine-grained parallelismFingerprint recognitionHeterogeneous computingLatent fingerprint identificationParallel processing
Sanchez-Fernandez, A. J., Romero, L. F., Peralta, D., Medina-Pérez, M. A., Saeys, Y., Herrera, F., & Tabik, S. (2020). Asynchronous Processing for Latent Fingerprint Identification on Heterogeneous CPU-GPU Systems. IEEE Access, 8, 124236-124253. [DOI: 10.1109/ACCESS.2020.3005476]
SponsorshipSpanish Government TIN2016-80920-R; Spanish Government TIN2016-80920-R; University of Malaga through the U-Smart-Drive project; I Plan Propio de Investigacion; Postdoctoral Fellowship of the Research Foundation of Flanders (FWO) 170303/12X1619N; ISAC Marylou Ingram Scholar; Spanish Government RYC-2015-18136
Latent fingerprint identification is one of the most essential identification procedures in criminal investigations. Addressing this task is challenging as (i) it requires analyzing massive databases in reasonable periods and (ii) it is commonly solved by combining different methods with very complex datadependencies, which make fully exploiting heterogeneous CPU-GPU systems very complex. Most efforts in this context focus on improving the accuracy of the approaches and neglect reducing the processing time. Indeed, the most accurate approach was designed for one single thread. This work introduces ALFI (Asynchronous processing for Latent Fingerprint Identification), the fastest methodology for latent fingerprint identification maintaining high accuracy. ALFI fully exploits all the resources of CPU-GPU systems using asynchronous processing and fine-coarse parallelism for analyzing massive databases. Our approach reduces idle times in processing and fully exploits the inherent parallelism of comparing latent fingerprints to fingerprint impressions. We analyzed the performance of ALFI on Linux and Windows operating systems using the well-known NIST/FVC databases. Experimental results reveal that ALFI is in average 22x faster than the state-of-the-art algorithm, reaching a value of 44.7x for the best-studied case.