Development of a signature image retrieval system

Pao Yue-kong Library Electronic Theses Database

Development of a signature image retrieval system

 

Author: Chu, Kwok-lung
Title: Development of a signature image retrieval system
Degree: M.Sc.
Year: 2002
Subject: Hong Kong Polytechnic University -- Dissertations
Digital signatures -- Design
Image processing -- Digital techniques
Department: Multi-disciplinary Studies
Dept. of Computing
Pages: xi, 64 leaves : ill. (some col.) ; 30 cm
Language: English
InnoPac Record: http://library.polyu.edu.hk/record=b1673141
URI: http://theses.lib.polyu.edu.hk/handle/200/2098
Abstract: Over twenty years, lots of researches have been done on automatic signature verification but less attention has been given to automatic signature retrieval.In this work, an off-line signature image retrieval system is proposed. It can be used to reduce forged signature problem by avoiding easily imitated signature to be enrolled in the signature reference database. There are ten categories of signature images used in this work, namely, s1 to sl0. Each category has 28 signatures, among which 12 are forged signatures while the other 16 are genuine ones. Each signature is pre-processed and feature extracted for the retrieval task. Both global features and local features are used for signature similarity computations. All the signatures within the same category are finally retrieved in descending order of similarity. Thinning is involved in the pre-processing stage and there are six features being extracted, namely, number of crossing points, signature aspect ratio (the width-to-height ratio), number of end points, number of turning points,number of smooth stroke segments from end points and crossing points, and number of non-smooth stroke segments from end points and crossing points. Each signature is sub-divided into certain local areas and the same features are captured in each area again. Each feature is given a weight in the similarity computation. The experimental results are measured in terms of precision for k=10 and k=15 number of returned items. The best overall performance of genuine signature retrieval is customer s1 with 98.13% and 94.58% precision for k=10 and k=15 respectively. The worst one is customer s6 with 71.25% and 62.92% respectively. In the overall performance of forged signature retrieval, the highest genuine signature retrieval rate is customer s2 with 36.67% and 42.22% precision for k=10 and k=15 respectively, showing that the signature set is the easiest one to be imitated. In the feature importance analysis, customer s7 is selected. The most important feature for the customer is local feature. Without it, the genuine signature retrieval performance dropped from 93.33% and 84.44% to 56.67% and 42.22% for k=10 and k=15 respectively. Customer s1 is selected to perform the sensitivity analysis. Noise adding, horizontal perspective view, vertical perspective view and rotation are introduced separately to the signatures for testing. The feature sets are found very sensitive to rotation of signature image and addition of pepper noise. Vertical perspective view has less effect on the feature set than horizontal perspective view.

Files in this item

Files Size Format
b16731414.pdf 6.671Mb PDF
Copyright Undertaking
As a bona fide Library user, I declare that:
  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

     

Quick Search

Browse

More Information