Signature recognition using wavelet transform

مقدم أطروحة جامعية

Abd, Raghad Salim

الجامعة

الجامعة التكنولوجية

الكلية

-

القسم الأكاديمي

قسم علوم الحاسوب

دولة الجامعة

العراق

الدرجة العلمية

دكتوراه

تاريخ الدرجة العلمية

2004

الملخص الإنجليزي

Lb the field of pattern recognition, offline handwritten signature reception is still an open problem.

Signature recognition can be cohered as a special case of pattern recognition.

Like in any pattern recognition problem, in signature recognition distinctive features can be extracted from a set of original signatures In this work, an approach to off-line handwritten signature recognition based on wavelet transform is presented.

This method possesses the benefits offered by the wavelet transform to extract a vector of values that describes the signature ™age features.

This approach employs feature vectors base on complex moments computed from the coefficients produced by ,he wavelet transform up.

to the fourth order Since the low order moments contain significant information about any simple object, they have been used as features for signature recognition.

In this work, wavelet transform is used with complex moments to keep : the order of the computed complex moments from signature ,mage low to minimize their sensitivity to noise, and to keep these moments values stable when the signature images are rotate Two types of measures used to classify the tested signatures they are : Euclidean minimum distance and city block measure In this work, we trained the proposed system in two ways: in the way only 70 original signature images were used as the training se.

A successful, identification rate of 93.3% was achieved.

In the first way only 70 original signature images were used as the training set.

Successful identification rate of 93.3 % wad achieved.

In the second way only 210 original signatures images were as the training set.

The training set is composed of 70 original signature images and 140 signature images resulting from rotating the 70 original, signatures images in +5, -5 degrees clockwise direction.

A successful identification rate of was achieved.

In both cases another 30 signature images were used ; testing of the proposed system.

Ring the work there were a number of parameters that had an Effect on the performance of the proposed system like the effect of ordinates normalization on the values of the calculated complex mints the effect of the complex moment values normalization on the recognition accuracy of the proposed system.

These effects are investigated and treated in this thesis.

It should be pointed out here that the programs that used in programming the work of this thesis was written in a Turbo Pascal version 7.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

الموضوعات

• معالجة البيانات
• برمجيات الحاسوب

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-306012