Breadcrumb

 
 

Automatic verification of the outputs of multiple classifiers for unconstrained handwritten numerals

Title:

Automatic verification of the outputs of multiple classifiers for unconstrained handwritten numerals

Tan, Jinna Michelle (2004) Automatic verification of the outputs of multiple classifiers for unconstrained handwritten numerals. Masters thesis, Concordia University.

[img]
Preview
PDF - Accepted Version
4Mb

Abstract

Recognition of unconstrained handwritten characters has gained considerable attention in different areas due to its many possible applications. Since the late 60's, research in this area has made impressive progress and many systems have been developed. Some of them achieved high recognition rates of 98% to 99%. However, these systems still misrecognize some patterns that are easily recognized by humans, thereby, decreasing the reliability of the system. For the purpose of satisfying the high demand of reliability in some practical applications such as bank cheque processing, we proposed a verifier based on structural features. The verifier is a prototype-based system. Different numbers of prototypes have been constructed for each digit according to the complexity of its structure. Prototypes are built using the structural primitives of a numeral and the relations among them. The local differences between some confusing pairs of numerals such as 4-9, 0-6 are also addressed in the prototypes. Three classifiers: SVM, LeNet5, and MQDF are used in the recognition stage of the proposed system. SVM is the primary classifier. Patterns rejected by the SVM are passed to the parallel combination of the three classifiers. The proposed verifier is then applied to the classification results. The proposed system yielded a reliability rate of 99.92% and a recognition rate of 96.50% on MNIST Database. The reliability increased from 99.06% to 99.92% after applying the verifier. Hence, we can conclude that the proposed system has successfully achieved high reliability while maintaining a reasonable recognition rate.

Divisions:Concordia University > Faculty of Engineering and Computer Science > Computer Science and Software Engineering
Item Type:Thesis (Masters)
Authors:Tan, Jinna Michelle
Pagination:x, 110 leaves : ill. ; 29 cm.
Institution:Concordia University
Degree Name:M. Comp. Sc.
Program:Computer Science
Date:2004
Thesis Supervisor(s):Suen, C. Y
ID Code:7985
Deposited By:Concordia University Libraries
Deposited On:18 Aug 2011 14:12
Last Modified:19 Aug 2011 03:59
Related URLs:
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Document Downloads

More statistics for this item...

Concordia University - Footer