Award Date

1-1-1997

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

First Committee Member

Tom Nartker

Second Committee Member

Junichi Kanai

Number of Pages

65

Abstract

Use of Neural Networks to Predict OCR Accuracy investigates issues in developing an artificial neural network (ANN) based system for prediction of OCR accuracy from the image of a page. This work extends the work of Blando and Gonzalez in the following ways: enlarging training data, proposing new features, comparing different ANN architectures, and introducing a cross-validation learning algorithm; The following experiments were performed: comparison of 14 dimension feature metrics and 7 dimension feature metrics, comparison of an ANN trained with and without cross-validation, comparison of different neural network architectures, comparison of prediction capability of neural network and linear regression, comparison of the prediction capability of neural network using 14 dimension feature metrics and linear regression using reject markers. The results show that neural network can outperform linear regression if properly trained, and that the new feature metrics provide improved predictive ability.

Keywords

Accuracy; Networks; Neural; OCR; Predict

Controlled Subject

Computer science; Artificial intelligence

File Format

pdf

File Size

1761.28 KB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Permissions

If you are the rightful copyright holder of this dissertation or thesis and wish to have the full text removed from Digital Scholarship@UNLV, please submit a request to digitalscholarship@unlv.edu and include clear identification of the work, preferably with URL.

Identifier

https://doi.org/10.25669/z8yl-cv5c


Share

COinS