Abstract

The thesis goal is to develop a computer system for hand printed digit recognition based on an investigation into various feature extractors and neural network strategies. Features such as subwindow pixel summation, moments, and orientation vectors will be among those investigated. Morphological thinning of characters prior to feature extraction will be used to assess the impact on network training and testing. Different strategies for implementing a multilayer perceptron neural network will be investigated. A high-level language called MatLab will be used for neural network algorithm development and quick prototyping. The feature extractors will be developed to operate on small (less than or equal to 256 bits) binary hand printed digits (0, 1, 2, 3, 4, 5, 6, 7, 8, 9).

Library of Congress Subject Headings

Neural networks (Computer science); Computer vision; Image processing--Digital techniques; Pattern recognition systems

Publication Date

10-1-1995

Document Type

Thesis

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Chang, Tony

Advisor/Committee Member

Anderson, Peter

Advisor/Committee Member

Salem, Edward

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: QA76.87.P564 1995

Campus

RIT – Main Campus

Download

Share

COinS