Project on Developing

An
Offline Handwritten Signature Recognition
Application

Presented By: Prasann Tamrakar (MU19MTCSE001)

Mtech. CSE, IIIrd SEM.
Guided By : Dr. Anurag Sharma
Professor, MATS University
OBJECTIVE
 The objective of signature verification systems is to
discriminate if a given signature is genuine, or a
forgery.
 The probability of two signatures made by the same
person being the same is very less. Many properties of
the signature may vary even when two signatures are
made by the same person. So, detecting a forgery
becomes a challenging task
 Here i want to present how the problem has been
handled in the past few decades, analyze the recent
advancements in the field, and the potential directions
for future research.
Introduction
 A signature is outlined as a uniquely written drawing that a person writes on any
document as an indication of identity. An individual uses it on a usual wish to sign a
check, a legal instrument, contract, etc. The matter arises when once somebody tries
to replicate it.
 Signature verification systems aim to automatically discriminate if the biometric
sample is indeed of a claimed individual. In other words, they are used to classify
query signatures as genuine or forgeries. Forgeries are commonly classified in three
types: random, simple and skilled (or simulated) forgeries.
 In the case of random forgeries, the forger has no information about the user or his
signature and uses his own signature instead. In this case, the forgery contains a
different semantic meaning than the genuine signatures from the user, presenting a
very different overall shape. In the case of simple forgeries, the forger has
knowledge of the user’s name, but not about the user’s signature. In this case, the
forgery may present more similarities to the genuine signature, in particular for
users that sign with their full name, or part of it. In skilled forgeries, the forger has
access for both the user’s name and signature, and often practices imitating the
user’s signature. This result in forgeries that have higher resemblance to the genuine
signature, and therefore are harder to detect.
Introduction
 Depending on the acquisition method, signature verification
systems are divided in two categories: online (dynamic) and
offline (static). In the online case, an acquisition device, such
as a digitizing table, is used to acquire the user’s signature.
The data is collected as a sequence over time, containing the
position of the pen, and in some cases including additional
information such as the pen inclination, pressure, etc. In
offline signature verification, the signature is acquired after
the writing process is completed. In this case, the signature is
represented as a digital image
 Deep Learning methods applied for handwritten signatures,
have demonstrated superior results in multiple benchmarks,
and is reviewed in the present work.
 Literature Review
 In the recent past, no work is reported for Offline signature verification using deep learning techniques. A
very few authors are working in this direction.
 Summary of previous relevant studies on signature verification.

Studies Key Features and Dataset Implementation method Results and Conclusion
Shahane Noise removal, preprocessing, Different thresholds used for matching The edge values used here
et al feature extraction, learning and to boost the general potency of the weren’t acceptable for all
verification modules system. It additionally includes the kinds of signatures because
verification of the account range and the values dissent from
quantity on the cheque using OCR and signature to signature of
finds out if the cheque is cleared or various persons.
bounced.
Fahmy Supported separate wave remodel It’s enforced on information of two Accuracy obtained is ninety
discrete wavelet transform (DWT) hundred signature having twenty real fifth for real signature. Here
feature extraction and uses a feed and twenty sure-handed forgery the dataset used for terribly
forward back propagation error signatures of 5 folks. little and over fitting
neural network for recognition prospects was high.

Zhu et al. It has a supervised learning Nursing approach to conjointly notice It is a complicated structural
framework for combining that signatures from document image framework.
complementary form data from can be extracted by structure study
totally different unsimilarity metrics detection across image scales.
via LDA.
Problem Statement: Challenges
 One of the main challenges for the signature verification task is
having a high intra-class variability. Compared to physical biometric
traits, such as fingerprint or iris, handwritten signatures from the
same user often show a large variability between samples. This
problem is illustrated in below Figure

Fig. Shows Superimposed examples of multiple signatures of the

same user. We can notice a high intra-class variability of the
signatures of the user.
Problem Statement: Challenges
 This issue is aggravated with the presence of low inter-class variability when we
consider skilled forgeries. These forgeries are made targeting a particular
individual, where a person often practices imitating the user’s signature. For this
reason, skilled forgeries tend to resemble genuine signatures to a great extent.
Another important challenge for training an automated signature verification system
is the presence of partial knowledge during training. In a realistic scenario, during
training we only have access to genuine signatures for the users enrolled to the
system
 During operations, however, we want the system not only to be able to accept
genuine signatures, but also to reject forgeries. This is a challenging task, since
during training a classifier has no information to learn what exactly distinguishes a
genuine signature and a forgery for the users enrolled in the system. Lastly, the
amount of data available for each user is often very limited in real applications.
During the enrolment phase, users are often required to supply only a few samples
of their signatures. In other words, even if there is a large number of users enrolled
to the system, a classifier needs to perform well for a new user, for whom only a
small set of samples are available.
Hypothesis
 Here we proposed a system where we normalize the signature image and system checks whether the
signature matches with original signature. System applies image pre-processing steps. Image is converted to
black and white image. Morphing technology is used in order to thin the image. By extracting black pixels,
the curve of the signature is recognized. X and Y co-ordinates of original image is extracted. New co-
ordinates are generated and signature is rotated by passing the new co-ordinates. After rotating the image,
Signature might go out of the boundary therefore we calculated the moving co-ordinates of x and y. Finally,
the image is cropped. This cropped image is compared with the theta value which is calculated during curve
calculation. System checks whether the theta value is matched. If the theta values are approximate. System
will display the result accordingly. System will provide result with accuracy 40%-60% depending on image
quality, image lighting and image background.
 Features
 System applies image pre-processing steps.
 Image is converted to black and white image.
 Morphing technology is used in order to thin the image.
 By extracting black pixels, the curve of the signature is recognized.
 X and y co-ordinates of original image is extracted.
 New co-ordinates are generated and signature is rotated by passing the new co-ordinates.
 System checks whether the theta value is matched.
 If the theta values are approximate. System will display the result accordingly.
Proposed Methodology
 To extract and procure correct feature results from the images, equipped
images should have an exact form, size, and alignment to strike a
comparison amongst all. Hence, numerous pre-processing techniques are
applied to realize the specified exactness within feature detection.
Methodology
 PREPROCESSING
As with most pattern recognition problems, preprocessing plays an important role in signature
verification. Signature images may present variations in terms of pen thickness, scale, rotation,
etc., even among authentic signatures of a person. Main preprocessing techniques are:
 Signature extraction

 Noise Removal

 Scalling

 Centralization

 Rotation
Methodology
 FEATURE EXTRACTION
Offline signature verification has been studied from many perspectives, yielding multiple alternatives for feature extraction.
Broadly speaking, the feature extraction techniques can be classified as Static or Pseudo-dynamic, where pseudodynamic
features attempt to recover dynamic information from the signature execution process (such as speed, pressure, etc.). Another
broad categorization of the feature extraction methods is between Global and Local features. Global features describe the
signature images as a whole - for example, features such as height, width of the signature, or in general feature extractors
that are applied to the entire signature image. In contrast, local features describe parts of the images, either by segmenting
the image (e.g. according to connected components) or most commonly by the dividing the image in a grid (of Cartesian or
polar coordinates), and applying feature extractors in each part of the image.
 A. Handcrafted feature extractors:
 Geometric Features:
 Graphometric features:
 Directional features
 Mathematical transformations:
 Shadow-code
 Texture features:
 Interest point matching
 Pseudo-dynamic features:

 B. Deep learning
In this approach, the system learns to compare two signatures, by learning a distance metric between them. The
signatures are processed using a feed forward neural network, where the bottom layers are shared among all users (i.e.
the same weights are used), and the last layer is specific to each individual, and specializes for the individual.
Conclusion and Future Scope
 The system successfully recognizes and identifies the signature holder accurately with the
forgery issue gift in it.
 A robust and reliable signature recognition and verification system with maximum accuracy
possible is very important for many purposes like enforcement, security management, and lots
of business processes. It can be used as an intermediate tool to authenticate several documents
like cheques, legal records, certificates, etc.
 The model gave encouraging results. Entirely different threshold values are used for feature
matching on testing and training vectors, which helped to boost the overall performance and
efficiency of the system.
 The planned system is highly economical in recognizing and sleuthing the forgeries at runtime
and therefore the responsibility of the system can be magnified by training the extracted
features on the Artificial Neural Networks by storing the extracted features. Negligible
misclassification or error is required in such sensitive applications although it’s at the cost of a
High Recognition Rate (HRR).
 Different aim is that the probability chance of forgery signature as if it’s a real one is zero. As
a future work, my aim is to increasing the resultant system accuracy by trying new and better
parameter coefficients that increases the deviation between real and forged signatures also i
want to make it online from offline so that we gets output at the real time.
Thank You