Obtención de las n mejores alternativas para clasificación de símbolos unicode

The Unicode character set has been increased in last years until grouping more than 100000 characters. We developed a classifier which can predict the n most probable solutions to a given handwritten character in a smaller Unicode set. Even with the size reduction we still have a classification prob...

Descripción completa

Detalles Bibliográficos
Autor: Vieco Pérez, Jesús
Tipo de recurso: tesis de maestría
Fecha de publicación:2017
País:España
Institución:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:riunet.upv.es:10251/86238
Acceso en línea:https://riunet.upv.es/handle/10251/86238
Access Level:acceso abierto
Palabra clave:Data generation
Neural network
Character recognition
Unicode
Reconocimiento de caracteres
Redes neuronales
Generación de datos
LENGUAJES Y SISTEMAS INFORMATICOS
Máster Universitario en Inteligencia Artificial, Reconocimiento de Formas e Imagen Digital-Màster Universitari en Intel·ligència Artificial, Reconeixement de Formes i Imatge Digital
Descripción
Sumario:The Unicode character set has been increased in last years until grouping more than 100000 characters. We developed a classifier which can predict the n most probable solutions to a given handwritten character in a smaller Unicode set. Even with the size reduction we still have a classification problem with a big number of classes (5488 in total) without any training sample. Before dealing with this problem we performed some experiments on the UJI PEN dataset. In these experiments we used two different data generation techniques, distortions and variational autoencoders as generative models. We tried feature extraction methods with both offline and online data. The generation along with the feature extraction was tested in several models of neural networks like convolutional networks or LSTM.