A review on signature detection and signature based document image retrieval

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 04 Issue: 03 | Mar-2015, Available @ http://www.ijret.org 251
A REVIEW ON SIGNATURE DETECTION AND SIGNATURE BASED
DOCUMENT IMAGE RETRIEVAL
Chinnu S Gupta1
, Umesh .D. Dixit2
1
M.Tech, E&CE dept, B.L.D.E.A’s College of Engineering and Technology, Vijayapur 586101, India
2
Asst. Prof in E&CE dept, B.L.D.E.A’s College of Engineering and Technology, Vijayapur 586101, India
Abstract
Transforming a paper document to its electronic version in a form suitable for efficient storage, retrieval, and interpretation
continues to be a challenging problem. Signature is an individualistic identification of a person. It is an authentic identification
because a signature cannot be copied by others. Signatures are a special case of handwriting subject to intra personal variation
and inter personal differences. To counter check fraud and forgery of handwritten signatures, Signature extraction from printed
text background and signature based document retrieval from a large dataset is necessary. A lot many techniques have been
implemented successfully for both signature extraction and signature based document retrieval. This paper present techniques
and methods evolved for signature extraction and signature based document retrieval.
Keywords: signature detection, signature extraction, Document image Retrieval, Query image retrieval
---------------------------------------------------------------------***--------------------------------------------------------------------
1. INTRODUCTION
A signature is an individualistic, unique, evidentiary entity.
It provides an important form of indexing that enables
effective image search and retrieval from large
heterogeneous document image collections. In this paper,
we surveyed different technique involved in retrieval system
that automatically detects, segments, and matches signatures
from document image with unconstrained layouts and
complex background. This would involve extracting all the
signatures from the documents and then performing a match
on these signatures. In searching complex documents, a task
of Relevance is relating a signature in a given document to
the closest matches within a database of document given a
database of signed document. The signature based document
retrieval has many of the applications and some of them are
listed below.
 For business documents.
 Government organizations.
 Digital libraries for online books, student thesis etc.
 Security requirements of document.
The main contributions of this paper are summarized as
follows. Firstly, in this paper we have provide general
framework and detailed survey of signature extraction and
signature based document retrieval. Secondly, we have
discussed about issues and challenges.
The paper is organized as follows. In section 2, we given the
general framework and in section 3, we review the related
work. The issues and challenges in section 4, in section 5,
we discuss the performance evaluation metrics. Section 6,
concludes the paper
2. GENERAL FRAMWORK
Fig 1.General framework
The figure shows general framework for signature based
document retrieval.
 Query image: The image containing signature is
called as query image.
 Signature extraction: It is the process of extracting
only the signature part of the document i.e.
separating signature and non-signature part of the
document.
 Preprocessing: The preprocessing phase is a
sequence of image transformations creating the best
possible input for feature extraction algorithms.
Preprocessing step may include filtering, RGB to
gray scale conversion and binarization.
 Feature extraction: Feature extraction is a set of
(usually) independent functions returning a
characteristic feature set for the input image. Features
can either be particular to the whole signature (global
features) or to a part of the signature (topological or
local features).
 Matching and Retrieval: Given a few available query
signature instances and a large database of document
containing signatures, the problem of signature
Query
Image
Signature
extraction
Preprocessing
Matchin
g and
Retrieval
Document
Retrieved
Database
Feature
extraction

_______________________________________________________________________________________
matching is to find the most similar signature
samples from the database. By constructing the list of
best matching signature documents, we effectively
retrieve the set of documents authorized or authored
by the same person.
3. RELATED WORK
Madasu, et al. [1] and Chalechale, et al. [2] used geometric
features including area, circularity, aspect ratio, size and
position to analyze segmented regions. These features are
compared based on certain parameters such as Manhattan
distance. The crop method objective is to locate rectangular
box around the signature which is the object of interest and
remove other objects outside this area, like signatures on
bank cheque. The crop method works by moving four
vectors from the four different directions (namely up, right,
bottom and left) towards the object of interest. Each vector
will mark the border of each side of the rectangular box. The
crop is then applied to area. Once the area approximation
has been done on the first model of the cheque. It is then
easy to extract a signature from a bank cheque and scan it.
The method failed to extract the signature as the signature
was not in the stipulated area hence the system could
approximate the area to be cropped.
Djeziri, et al. [3] dealt with the signature extraction problem
by an approach that was to mimic the human visual
perception. They introduced the filiformity as a criterion for
the curvature characteristics of handwritten signatures.
Filiformity is defined for two topological measures for
binary objects which also includes gray level images. It
differentiates the contour lines of the signatures from the
handwritten lines which are being isolated. The process of
isolating is also used to provide measure for local values
regarding the whole image. This process fails when other
filiformity objects are present in the document.
Madasu, et al. [4] tried to crop the image segment by
estimating the area in which the signature lies using a
sliding window. They then analyzed the local entropy
derived from the pixel-based density of the region to decide
its being signature or not. This approach disregarded the
noise and therefore high-density regions are reported as
signatures incorrectly. This segmentation does not need any
a priori information about the data field and features.
Ritesh Banka, et al. [5] proposed the method of extracting
signature from cluttered background which is scale invariant
technique using the efficiency of symmetric and loop
features that are common in signatures for extracting the
standard signatures from other documents. Signature is a
handwritten text hence this technique can be used because
signatures in English contain most of the characters
exhibiting symmetry in vertical, horizontal or diagonal
directions. The self printed characters from the documents,
as a few papers, lack of proper self symmetry and are
misclassified as handwritten text at the first level. The
misclassified printed characters in the first level contain a
single symmetric loop, which discriminates them from hand
written elements. The algorithm had two domains character
level and document level. For poor quality document where
the numbers of characters joint and broken are very high, the
algorithm efficiency decreases.
R.Jayadeyan, et al. [6] considered the images of bank
document a method has been proposed for the extraction of
signatures, signature localization is done using variance,
signature block extraction is done using entropy,
normalization are used for desired mean and variance, a
hidden Markov model (HMM).Horizontal and vertical
projection, right and left envelope, top and bottom envelope,
horizontal and vertical size variations of the binary format of
the genuine signatures are considered as the feature set of an
individual.
Sha’ashua, et al. [7] introduced the concept of multi-scale
saliency feature for signature detection that defines signature
characteristics by identifying salient structure by grouping
contours. They defined a saliency function which increased
with the curve length and decreased with the curvature when
totally squared. This was done by locally connected
network. Their approach was confined to rigid objects
having 1-D contour. This can be applied to handwritten
notes collected on a tablet pc online since trajectories of the
pen are available the framework used is general as it does
not embed unique assumption on local feature of the object
for example stroke level features for signatures. Hence it is
robust against any changes in shape based object detection
problems and is applicable to different languages.
A three-stage procedure was proposed by Mandal, et al. [8]
to extract signatures. First an algorithm is used to locate the
Signature in the document using word-level feature
extraction. Second stage separates signature strokes that
overlap with the printed text. Final stage uses conditional
random field minimization energy concept with skeleton
analysis to classify real signature strokes. Improper
segmentation of strokes causes error in separating signature
from printed text.
Bassam, et al. [9] used the method for extracting signature
from image on document is the base proposed auto cropping
method. This method improves the performance of security
system based on signature image as possible the region of
interest of the used image for the biometric system and it
also reduces the time cost associated with signature. Auto
cropping is the fast procedure to extract the Region of
interest (ROI). In this method they used image segmentation
and extraction. This cropped signature has no garbage
region it crops only the ROI of signature image. This takes
less processing time then the original signature image. The
performance of this method shows through its speed time,
and keeping the content information of the signature object
without losing of any pixel of image.
Esteban, et al. [10] considered two main stages of extraction
one being comparing distribution of stroke in model
signature against distribution of stroke in the original
signature using a different approaches extraction being the
main concern accumulative evidence technique is used, and

_______________________________________________________________________________________
they achieved a high accuracy with this approach. Their
assumption is not usually true since the bank application
forms are often filled by new customers and this process is
not followed by verification but an archiving step.
Tang, et al. [11] proposed a technique to automatically gain
knowledge from various types of documents because
knowledge acquisition is difficult to obtain. It was proposed
to gain automatic knowledge acquisition in document
images by analyzing the geometric structure and logical
structural images. These two structures play an important
role to process the knowledge acquisition. The limitations of
these techniques are to find which is right in transforming a
geometric structure into a logical structure. Another
difficulty is to find the correct rules involved in obtaining
knowledge from different documents.
Liu, et al. [12] presented an approach to image-based form
document retrieval and they proposed a similarity measure
for forms that is insensitive to translation, scaling, moderate
skew, and image quality fluctuations, and developed a
prototype form retrieval system based on their proposed
similarity measure.
G.Zhu, et al. [13] proposed the method for signature based
document retrieval performance using different shape
representation like salient contour computed by detection
and segmentation of the skeleton that are extracted directly
from labeled signature region of database .The thinning
algorithm shows sensitivity to structural variations due to
noise and neighboring stroke salient contours give a globally
consistent representation of structurally important shape
feature. This retrieval technique has been successful on the
Maryland Arabic dataset in which background handwritten
as well as signature are closely spaced.
Chalechale, et al. [14], proposed signature based
decomposition and retrieval of document images, and he
investigated Arabic/Persian signature recognition and
retrieval. For the conduction of retrieval automatic links
were considered between feature vector of the signature and
the document containing the image using file names. The
retrieval performance was measured by average normalized
retrieval rate. The ANMRR indicates better performance in
comparison with line segmentation distribution method.
A.chalechale, et al. [15] proposed new method for document
image decomposition based on connected component
analysis and geometric properties of labeled region. The
signature is detected and extracted by spatial partitioning by
accumulating pixels and by using magnitude and Fourier
transform they achieved rotation invariance. The main
objective is signature extraction from the original image and
converted into a compact feature vector that supports
measuring signature similarities.
Srihari, et al. [16], proposed a document image retrieval
using signature as queries. Using global shape binary feature
vector, a normalized correlation similarity measure for
signature matching is done. The objective is to retrieve the
closest matching signature obtained from dataset by the
same signature person to remove the printed text from the
signature image. An image enhancement procedure using
chain code information is used. The technique gave
promising result for group based and non grouped base
retrieval with accuracy and precision.
H.Srinivasan, et al. [17] used conditional random fields and
proposed the method for signature based retrieval. In this
they retrieved the document from database using signature
image as query. Isolating the different contents present in
the document and with the help of CRF in extracting
signature from complex document. This method presents the
signature retrieval strategy using document indexing and
retrieval. Indexing is done by using a model based on CRF.
SVM is also supported by this technique, and the CRF is
used to label each patch and identified using the labels of the
neighboring patches. Document retrieval is performed by
using matching algorithm to compare the query with the
signature.
Guangya Zhu, et. Al. [18], proposed a signature-based
document image retrieval system that automatically detects,
segments, and matches signatures from document images
with unconstrained layout and complex backgrounds
.Signature is treated as a non rigid shape and represented by
discrete set of 2-D point to most commonly sited measures
for retrieval R-precision and average precision. The R-
precision emphasizes the ranking among retrieve
documents. Extensive experimental and field test give
excellent performance for document retrieval
4. ISSUES AND CHALLENGES
The issues and challenges for signature extraction and
retrieval are listed below.
 Signature with low resolution in documents make
difficult for detection and segmentation.
 The background of document differs from each other.
 The computer vision faces an important problem of
Detecting, segmenting and matching deformable
objects such as signature.
 Documents are subjected to restricted processing
time due to urgency of applications. Therefore the
detection and retrieval time must be fast.
 The handwritten characters and auxiliary lines
contained in the document overlap and resemble
signatures.
5. PERFORMANCES AND EVALUATION
METRICS
 Accuracy: It is the percentage of number of correctly
detected signatures to the number of groundtruthed
signatures from the document image.
Accuracy =
a
a h

_______________________________________________________________________________________
 Precision: This refers to the percentage of retrieval of
document images that are relevant to query [19].
Precision (N) =Rn / N
Where N is number of retrieval and Rn is number of
relevant matches among retrievals.
 Recall: This refers to the percentage of all the
relevant document images in the search database
which are retrieved.
Re call (N) =Rn /M
Where M is the total number of relevant matches in
the database Rn is the number of relevant matches
among retrievals.
A strategy for evaluating document image retrieval system
involves the following techniques. These techniques have
been partially successful in signature based document
retrieval. The technique involves extraction of signature
from certain documents by different methods. Matching of
signature is done by several method and retrieval by known
successful techniques.
6. CONCLUSION
This paper provides detailed survey of techniques, as it
overcomes the difficulty and challenges faced during
extraction of signatures from the window of different
background documents, cheque etc. This paper provides the
overview of methods to apply according to the type of the
background document and signature window. These newly
developed methods are going to give a high rate of
recognition. The performance of the most renowned
methods provided the higher signature recognition rate with
a greater accuracy in extraction and recognition of the
signatures and signature based documents. It also highlights
issues and challenges in this area.
REFERENCES
[1]. V K Madasu ,M H M Yusof M Hanmandilu K
K b ka,”Automic Extraction Of Signatures From Bank
Cheques And Other Documents”, DICTA’03 2003
[2]. A.Chalechale, G.Naghdya, P.Permaratne, A.Martins ”
Document Image Analysis and Verification using Cursive
S a ”, IEEE I a a C M m a
and Expo, 2004.
[3]. S.Djeziri,F.Nouboud,R.Plamondon, “Extraction Of
Signatures From Check Background Based On A Filiformity
Criterion”, IEEE Trans. Image Process. 7(10), 1425–1438,
1998.
[4]. V.K. Madasu, B C L v ,” Automatic Segmentation
and R O Ba k Ch q F ” Digital Image
Compute. Tech. Appl., 80(1): 33–40. 2005
[5]. R h Ba ka, Fa ha N bakh h,”Exraction Of
Signature &Handwritten Region From Official Binary
D m Ima ”,september, 2008
[6]. R Ja a va a ,” Variance based extraction and
hidden Markov model based verification of signatures
p ba k h q ”, International Conference on
Computation Intelligence and Multimedia Applications.
2007
[7]. G. Zhu, Y. Zheng, D. Doermann, S. Jeager, “Multi-scale
Structural Saliency for Signature D ”, IEEE
Conference on Computer Vision and Pattern Recognition.
2007
[8]. R. Mandal, P.P. Roy, U.Pal, “Signature Segmentation
from Machine Printed Documents using Conditional
Ra m F ”, International Conference on Document
Analysis and Recognition. 2011.
[9]. BassamAl-Mahadeen and Mokhled S,Islam
H A Ta aw h, ”Signature Region of interest using auto
pp ”, IJCSI,march 2010
[10]. J.L.Esteban, J.F. Vé z, Á Sá h z, ”Off-Line
Handwritten Signature Detection By Analysis Of Evidence
A m a ”, IJDAR, 15:359–368. 2012.
[11]. Y.Ta , C D Ya, a C Y S , ”Document
Processing for Automatic Knowledge Acquisition. IEEE
T a K w a Da a E ”, vol.6, no.1, pp.3-21,1994
[12]. J. Liu and A K Ja ”Imaged-Based Form Document
R va ”, Pattern Recognition, vol.33, no.3, pp.503-513,
2000.
[13]. Guangyu Zhu, Yefeg Zheng, and David
D ma ,”Signature Based Document Image Retrieval”,
ECCV, 2008, Part III, LNCS 5304, pp.752-765, 2000.
[14]. Abdul ah Cha ha , G hah Na h , ”Signature
Based Document Retrieval. Faculty Of Information-
Papers”, University of Wollongong.
[15]. Sargur N. Srihari, Shravya Shetty, Gady Agam and
Ophir Frieder .2006. Document Image Retrieval Using
Signature as Queries. In Proceedings of the Second
International Conference on Document Image Analysis for
Libraries (DIAL’06)
[16]. H. Srinivasan and sargur Sridhar,” Signature-Based
Retrieval Of Scanned Document Using Conditional Random
F ”,2009.
[17]. I kha , Ha a O ,”D Ha w
Signature In Scanned Documents “, February, 2014
[18]. M B K ka a M S Sh h ka , “D m
Ima R va : A Ov v w”, I a a J a
Computer Applications, vol. 1, no. 7, (2010), pp. 114-119.
BIOGRAPHIES
Miss. Chinnu s. Gupta has completed her
Bachelor of Engineering in Visvesvaraya
Technological University, Belgaum.
Currently pursuing Master in Technology,
from the same university. Area of interest
in Image Processing.
Mr. Umesh .D. Dixit has completed his BE
And M.Tech from Visvesvaraya
Technological University, Belgaum. He is
working as Asst.Prof.in the department of
E&C, BLDEA’ CET, V ja ap , 11
Years. Currently he is pursuing PhD in
Visvesvaraya Technological University,
Belgaum.

A review on signature detection and signature based document image retrieval

More Related Content

What's hot (19)

Similar to A review on signature detection and signature based document image retrieval (20)

More from eSAT Journals (20)

Recently uploaded (20)

A review on signature detection and signature based document image retrieval