Publication: Aceleración con GPU de algoritmos de matching para reconocimiento biométrico mediante patrón de iris
Loading...
Identifiers
Publication date
2014-09
Defense date
2014-10-03
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
El siguiente trabajo consiste en la implementación del algoritmo de matching para reconocimiento biométrico mediante patrón de iris. El algoritmo de matching permite la comparación de patrones de iris para la identificación o verificación de individuos.
Por lo general, las técnicas de reconocimiento biométrico requieren de un elevado coste computacional debido a que se procesan elevadas cantidades de datos. En este trabajo se acelera el mencionado algoritmo mediante GPU (Graphics Processing Unit) aprovechando la potencia de estas presentan. Las tarjetas gráficas de NVIDIA permiten la programación de propósito general (GPGPU) para desarrollar aplicaciones que usualmente se procesan en la CPU.
La aceleración mediante GPU consiste en la paralelización del algoritmo realizándose así varias tareas simultáneamente con el objetivo de aumentar notablemente el rendimiento de la aplicación. Con el objetivo de estudiar el rendimiento de la aplicación acelerada sobre la GPU respecto a la secuencial sobre la CPU, se implementa el algoritmo de las dos maneras exponiéndose una comparativa entre ambos.
Los lenguajes de programación empleados en el desarrollo del trabajo son C para la implementación sobre la CPU y C con extensión CUDA para la implementación paralela sobre la GPU.
Para poder entender los conceptos del proyecto, se realiza una introducción teórica sobre la biometría y el reconocimiento de iris destacando el algoritmo de matching, así como una introducción sobre la computación paralela centrándose en la programación sobre las GPUs.
The following work consists of implementing the matching algorithm for biometrics recognition through the iris pattern. The matching algorithm allows the comparisons between iris patterns for individuals identification or verification. Generally, biometrics recognition techniques require a high computing cost due to the high amount of data being processed. In this work the mentioned algorithm is sped up through the GPU (Graphics Processing Unit) using their power. Video cards by NVIDIA allow the general purpose computing (GPGPU) to develop applications that are usually processed in the CPU. The acceleration through GPU lies in parallelizing the algorithm processing several tasks simultaneously. The aim of this is to increase the application performance significantly. In order to study the improvement in the performance of the sped up application on the GPU with regard to the sequential application on the CPU, the algorithm will be implemented in both ways exposing a comparison between them. The programming languages used in the development of the work are C for the implementation on the CPU and CUDA for the parallel implementation on the GPU. In order to be able to understand the concepts of this work, it's required a theoretical introduction about biometrics and iris recognition standing out the matching algorithm, as well as an introduction about parallel computing focusing in programming on GPUs.
The following work consists of implementing the matching algorithm for biometrics recognition through the iris pattern. The matching algorithm allows the comparisons between iris patterns for individuals identification or verification. Generally, biometrics recognition techniques require a high computing cost due to the high amount of data being processed. In this work the mentioned algorithm is sped up through the GPU (Graphics Processing Unit) using their power. Video cards by NVIDIA allow the general purpose computing (GPGPU) to develop applications that are usually processed in the CPU. The acceleration through GPU lies in parallelizing the algorithm processing several tasks simultaneously. The aim of this is to increase the application performance significantly. In order to study the improvement in the performance of the sped up application on the GPU with regard to the sequential application on the CPU, the algorithm will be implemented in both ways exposing a comparison between them. The programming languages used in the development of the work are C for the implementation on the CPU and CUDA for the parallel implementation on the GPU. In order to be able to understand the concepts of this work, it's required a theoretical introduction about biometrics and iris recognition standing out the matching algorithm, as well as an introduction about parallel computing focusing in programming on GPUs.
Description
Keywords
Biometría, CUDA, Distancia de Hamming, Algoritmo de matching