Sie sind auf Seite 1von 19

Maharashtra Institute of Technology, Aurangabad

Department of Computer Science and Engineering


Signature Recognition Using Machine
Learning

Presentor-Manali Tawar
Guided by-S.R.Chaudhary
Objectives
•Identification
of signature owner
•Decision whether the signature is genuine or
forger

Necessity
• Security
• Authentication
• Signature as a ‘seal of
approval’

Signature Recognition Using Machine Learning


Area of applications-
• Finance
•Insurance
•Real estate
•Health
•Telecom

Technical Area-
•Image Processing
•Machine Learning

Signature Recognition Using Machine Learning


Problem Statement
Recognizing signature by considering the assumptions that
due to the use of different pens/stylus will not change the
owner of signature, the fact that no two signatures of the
same person are exactly same and any mark on the paper or
any such element should not affect while recognizing the
signature.

Signature Recognition Using Machine learning


Introduction
1) Online 2) Offline

Signature

Genuine Forgery

Random Skilled Simple


forgery forgery forgery

Signature Recognition Using Machine Learnng


Literature Surveys

Handwritten Signature Recognition And


Verification
Title Author Dataset Method Advant Drawba Future
/Classifi age cks scope
er
1.An Offline Ioana 84 Naïve Accuracy Too many Systems
System for Barbantan individual Bayes of 91.04% redundant capabilitie
Handwritte a and signature And attributes s of
n Signature
Recognition
Raluca with 20 Multilayer biases the detecting
IEEE 2009 Barca samples of Perceptio model forged
each n signature.

2. On-line FRR-0.0% FAR-<0.1


handwritten Dariusz Z. 41 % Investing
signature Lejtman individual other
verification
using
and Susan signature architectu
wavelets E. George res for the
and Backpropa NN with
back- gation view to
propagation algorithm reducing
neural the
networks number of
IEEE 2001
weights
making
NN
compact
for use in
Title Author Datase Metho Adv Drawb Future
t d acks scope
3.Deep Bernardet GPDS Boltzmann 84.07% To
Learning e Ribeiro1, synthetic Machine accuracy perform
Networks Ivo dataset and deep extensive
for Off- Gonçalves neural study to
Line 1, and network cope with
Handwritt Sérgio the
en Santos1, millions of
Signature and parameter
Recognitio Alexander s that
n Kovacec2 need to be
Springer adjusted
2011 with the
EER 5%
Database More use of
for 5
made of Preprocess references Graphics
reference
400 ing:segmen more Processing
signatures
Alessandra signatures ting by accuracy Unit.
and EER
4. A hybrid Zimmer, with 20 stroke Learning
1% for 10
On/Off Lee Luan samples of complexity process
reference
line Ling each with the
signatures
Handwritt introducti
en on of new
signature metrics
verification and the
system Handwritten inclusion
Signature Recognition And
IEEE 2003 Verificationof
Title Author Datase Metho Advant Disadv Future
t d ages antages scope

5.An Yasmine CEDAR SVM,Grap Accuracy Classifier is Try to


improved Serdouk, dataset h rate 93.9% related to extend the
Artificail Hassiba matching,H user AIRS-SVM
Immune Nemmour armonic defined for
Recognitio and Youcef distance,M parameters multiclass
n System Chibani LP,Adaptiv classificatio
for Offline e feature n such as
Handwritt thresholdin signature
en g recognitio
Signature n
Verification
IEEE 2015.

Handwritten Signature Recognition And


Verification
Procedure
1. Acquire the signature image
2. Image-preprocessing
3. Extract the various features
4. Use these features to train the system using
KNN algorithm
5. Test the signature
6. Take decision as original or forgery
Block diagram

Import the Dataset

Creating CNN Layers

Train the Model

Classify the output

Genuine/Forgery

Signature Recognition Using Machine Learning


Image preprocessing

•Resizeimage to 100*200 pixel


•Convert image into black and white
Feature Extraction

If we are to compare two sketches; there should be


at least one measurement on which to base this
comparison. The main function is to generate features
which can be used as comparison measurements.
This will be implemented using Histogram of
Oriented Gradient.
Training and Testing
•To store the extracted features of the dataset and
use this as a reference when classifying signature
using KNN.
Proposed
System
Design Signature

Yes
No
Input 10 Already
Test Signature
signature Registered

Train
data Preprocessing
using Process
NN in NN

Save Success
trained Accept
Verified
data as
reference
Fail
Database(
stored in Reject
memory) Handwritten Signature Recognition And
Verification
Methodology/Algorithms

Feature-Extraction-
Histogram of Oriented Gradient
Classifier-
K-Nearest Neighbor Classifier
Histogram of Oriented Gradient
KNN Classifier
Handwritten Signature Recognition And
Verification

Das könnte Ihnen auch gefallen