Publication: Sistema basado en redes neuronales para el reconocimiento de dígitos manuscritos
Loading...
Identifiers
Publication date
2012-02
Defense date
2012-02-16
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Este proyecto surge de la necesidad de mejorar el módulo de reconocimiento de dígitos del sistema Eyegrade. Este sistema realiza la corrección automática de exámenes tipo test mediante una cámara web. El modulo a mejorar sirve para identificar cada examen con un número de identificación del que dispone cada alumno. De esta manera la corrección es totalmente automática, sencilla de realizar e independiente de la acción humana. En este trabajo se han estudiado diferentes alternativas por las que resolver este problema, usando la Inteligencia Computacional , nuevas técnicas que cada vez van tomando más relevancia en el mercado para automatizar decisiones que suele hacer el ser humano. Después de estudiar y evaluar las diferentes alternativas para el problema planteado y las diferentes opciones que había para implementarlas, se decidió optar por las redes neuronales. Se pensó en hacer primero un prototipo de la solución en un entorno que nos facilitara su implementación, ya que era una gran ventaja. Se decidió que fuera MATLAB ©. Una vez se tuvieron todos los parámetros, tanto de las imágenes como de la red neuronal, se implementó el nuevo modulo en el lenguaje de Python,y así poderlo integrar con el sistema del cual surge esta necesidad. Las imágenes utilizadas para los experimentos, tanto en el prototipo de la solución como en el sistema nal, eran reales y extraídas de exámenes de la Universidad Carlos III de Madrid, con las que se pudieron extraer los resultados nales. ___________________________________________________________________________________________________________________________
This project emerges from the necessity to improve the digit recognition module Eyegrade system. This system executes the automatic correction for multiple-choice exams through a web-cam. The module at improving is useful to identify each exam with an ID number which belongs to each student. In this way, the correction is fully automatic, easy to perform and independently of human action. In this work, diff erent alternatives have been studied to solve this problem, using soft computing , new techniques which are becoming more relevant in the market to automate decisions that often are done by humans. After studying and evaluating the diff erent alternatives for the problem and the various options that had to be implemented, the chosen option was the neural networks. Firstly, a prototype of the solution was made in an environment that we provide implementation, as it was a great advantage. MATLAB©was the chosen one. When all the parameters were taken not only the images but the neural network as well, it was implemented in the Python language, for integrating with the system. Both the prototype of the solution and the final system, were performed with real images from exams of the Carlos III University of Madrid. The fi nal results were obtained with the mentioned images.
This project emerges from the necessity to improve the digit recognition module Eyegrade system. This system executes the automatic correction for multiple-choice exams through a web-cam. The module at improving is useful to identify each exam with an ID number which belongs to each student. In this way, the correction is fully automatic, easy to perform and independently of human action. In this work, diff erent alternatives have been studied to solve this problem, using soft computing , new techniques which are becoming more relevant in the market to automate decisions that often are done by humans. After studying and evaluating the diff erent alternatives for the problem and the various options that had to be implemented, the chosen option was the neural networks. Firstly, a prototype of the solution was made in an environment that we provide implementation, as it was a great advantage. MATLAB©was the chosen one. When all the parameters were taken not only the images but the neural network as well, it was implemented in the Python language, for integrating with the system. Both the prototype of the solution and the final system, were performed with real images from exams of the Carlos III University of Madrid. The fi nal results were obtained with the mentioned images.
Description
Keywords
Dispositivos ópticos de reconocimiento de caracteres, Redes neuronales