Javascript must be enabled for the correct page display

Context-enhanced text recognition using semantic vector space models

Gankema, T. (2016) Context-enhanced text recognition using semantic vector space models. Master's Thesis / Essay, Artificial Intelligence.

masterAI_2016_TGankema.pdf - Published Version

Download (1MB) | Preview
[img] Text
Toestemming.pdf - Other
Restricted to Backend only

Download (90kB)


Contextual information is important to correctly recognize text. Several methods exist to use contextual information in text recognition systems, most notably statistical language models. However, these models have the disadvantage that they can not generalize over similar sequences of words, because they treat words as atomic units. Moreover has previous work failed to find a satisfiable solution to add semantic context information to text recognition systems. Several attempts have been made, however these often require a lot of manual work and do not scale to large vocabularies. In recent years, with the rise of deep learning techniques, substantial progress has been made to build semantic representations of words based on large text corpora with unsupervised methods. We propose a way to use these so called semantic vector space models to add semantic context information to text recognition systems. Two different neural network architectures are implemented and experiments show that they are able to make useful predictions about the meaning of unknown words based on the context in which these words occur. Experiments with a simulated text recognition system show that the proposed methods perform better than N-gram models when trained on the same data. The proposed methods can improve current text recognition systems, because they add semantic information, scale to large vocabularies and generalize over semantically similar words.

Item Type: Thesis (Master's Thesis / Essay)
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 15 Feb 2018 08:25
Last Modified: 15 Feb 2018 08:25

Actions (login required)

View Item View Item