OFFLINE HANDWRITTEN SIGNATURE RECOGNITION USING EDGE HINGE AND EDGE EXTRACTION TECHNIQUES AND MANHATTAN DISTANCE CLASSIFIER

http://www.iaeme.com/IJECET/index.asp 1 editor@iaeme.com
International Journal of Electronics and Communication Engineering and Technology (IJECET)
Volume 8, Issue 1, January - February 2017, pp. 01–10, Article ID: IJECET_08_01_001
Available online at
http://www.iaeme.com/IJECET/issues.asp?JType=IJECET&VType=8&IType=1
ISSN Print: 0976-6464 and ISSN Online: 0976-6472
© IAEME Publication
OFFLINE HANDWRITTEN SIGNATURE
RECOGNITION USING EDGE HINGE AND EDGE
EXTRACTION TECHNIQUES AND MANHATTAN
DISTANCE CLASSIFIER
Padmajadevi G
Associate Professor, MCE, Hassan, Karnataka, India
Dr. Aprameya K.S
Research Guide, UBT, Davanageri, Karnataka, India
ABSTRACT
In this paper, another way to deal with distinguish, the issue that emerges in offiline manually
written signature recongition as been examined. As opposed to numerous current framework, an
endeavor made utilizing Edgehinge and Edge Extraction method. To finish this assignment, as an
initial step the mark test is preprocessed and standardized to perform content distinguishing proof
effectively. The second step removing the data and educational components occured amid this
period. At long last the Signature Sampleization decission was done in view of Data Acquisition
and Preprocessing, Feature extraction. Edgehinge dissemination is an element that describes the
adjustments in heading of a script stroke in the specimen of signature. The edgehinge
appropriation is extricated by method for a windowpane that is slid over an edge-identified double
picture. At whatever point the mid pixel of the window is on, the two edge pieces (i.e. associated
arrangements of pixels) rising up out of this mid pixel are measured. Their headings are measured
and put away as sets. A joint likelihood appropriation is acquired from an expansive specimen of
such matches. Regardless of ceaseless exertion, offiline signature recognizable proof remains a
testing issue, because of various methodologies use distinctive assortments of elements, having
diverse. Subsequently, our study will concentrate on acknowledgment in light of highlight
determination to rearrange highlights removing assignment, enhance Signature Sampleization
framework many-sided quality, lessen running time and enhance the order precision.
Key words: recognition, signature
Cite this Article: Padmajadevi G and Dr. Aprameya K.S, Offline Handwritten Signature
Recognition Using Edge hinge and Edge extraction Techniques and Manhattan Distance Classifier,
International Journal of Electronics and Communication Engineering and Technology, 8(1), 2017,
pp. 01–10.
http://www.iaeme.com/IJECET/issues.asp?JType=IJECET&VType=8&IType=1

Padmajadevi G and Dr. Aprameya K.S
1. INTRODUCTION
The issue of manually written mark cheque has been well established in examination group most likely on
account of its different valuable applications in everyday life, for example, cheque extortion recognition,
managing an account exchanges, programmed reserve exchanges, individual recognizable proof, official
courtroom. The online-based mark confirmation frameworks have been appeared to give a somewhat high
accuracy rate since they take quite a bit of advantage of element highlights like increasing speed, speed,
the request of strokes, weight, and constrain data. This dynamic data is gained straightforwardly amid the
written work process utilizing extraordinary electric marking gadgets. In the opposite, the disconnected
from the net based framework gets as the contribution of dark scale on the other hand twofold pictures
which are procured by scanners after the marking procedure has as of now been finished. As no dynamic
data is caught on the mark pictures, the issue of disconnected from the net mark confirmation turns out to
be a great deal more confused. Regardless, the offline based frameworks require no uncommon gadgets for
information securing, and the greater part of the marks are, in day by day life, displayed in the types of
writings, papers or records. In this manner, it draws much of enthusiasm from exploration group to outline
hearty logged off mark cheque frameworks. The issue of programmed logged offline/online written by
hand signature cheque has been well established in examination group likely on account of its different
helpful applications in everyda life, for example, cheque misrepresentation discovery, saving money
exchanges, programmed reserve exchanges, individual ID, courtroom. The online-based mark confirmation
frameworks have been appeared to give a fairly high exactness rate since they take quite a bit of advantage
of element highlights like speeding up, speed, the request of strokes, weight, and compel data. This
dynamic data is procured straightforwardly amid the written work process utilizing unique electric marking
gadgets. In the opposite, the logged off based framework gets as the contribution of dark scale on the other
hand twofold pictures which are obtained by scanners after the marking procedure has as of now been
finished. As no dynamic data is caught on the mark pictures, the issue of disconnected from the net mark
confirmation turns out to be significantly more convoluted. In any case, the offline based frameworks
require no uncommon gadgets for information procurement, and a large portion of the marks are, in
everyday life, introduced in the types of writings, papers or records. Hence, it draws much of enthusiasm
from exploration group to plan hearty disconnected from the net mark cheque frameworks. The vast
majority of the current frameworks take after a general structure of a mark verifier comprising of three
fundamental stages: prepreparing, highlight extraction and grouping [8]. In the first stage, the
commonplace undertakings are mark extraction, commotion expulsion, skew revision, incline adjustment,
size standardization, binarization, skeletonization, and so on. In the second stage, two sorts of elements are
regularly utilized including nearby components (e.g., matrix based elements, pixel-based highlights, nearby
thickness, width/stature proportions, stroke introductions, crossing-focuses) and worldwide highlights
(Fourier-, Irregular, Wavelet-based elements). In the arrangement stage, a classifier could be chosen as a
separation based capacity (e.g., Euclidean, DTW, Mahalanobis) on the other hand a machine-learning-
based model (e.g., GMM,SVM, Well). The yield of this stage is frequently constrained to a twofold choice
(i.e., hard choice). While this paired classification based methodology is somewhat basic in the writing,
late works [3, 12, 16] have demonstrated that such an arrangement of settling on a hard choice is not
reasonable by and by in view of the accompanying reasons: It is not the charge of the framework to settle
on a definite conclusion on creation yet the charge of the client (e.g., a courtroom); The framework has no
earlier data about the present situation (i.e., the earlier likelihood of origin and the expense of making
incorrectly/genuine choice); The errand of selecting a decent limit for settling on choice is not trifling;
Expansion of the computational challenges on the off chance that one needs to consolidate the yield of
various biometric frameworks together (e.g., DNA-, Face-,Unique mark , Discourse based frameworks).
These certainties persuaded us to build up a disconnected from the net mark cheque framework whose
fundamental target is to give the client two yields: a proof score assessed from the info designs and the
level of assurance of finding that score. This kind of methodology alludes to the structure of settling on
delicate choice, as said in [16], and the framework is considered as a criminological penmanship master
that he looks at the info designs (e.g., commonly including one doubted signature and an arrangement of

Offline Handwritten Signature Recognition Using Edge hinge and Edge extraction Techniques and Manhattan
Distance Classifier
real marks from a reference author) by means of two inverse speculations H0 and H1 expressed as takes
after.
Figure 1 Original Speciman
Figure 2 Forgery Speciman
1) Speculation H0: The addressed mark is a bona fide one
2) Speculation H1: The addressed mark is a fabrication one
2. RELATED WORK
In this area, we quickly survey the primary methodologies for logged off written by hand signature cheque
and primarily concentrate on the phases of highlight extraction and order. These methodologies can be
sorted into two classes:
Worldwide and nearby methodologies. For the previous approaches, the creators in [8] introduced two
techniques devoted to highlight extraction. In the first place technique tracks the positional variety of the
vertical projection profiles of the mark and afterward utilizes the dynamic time wrapping (DTW)
procedure for coordinating. The second strategy measures the positional variety of individual strokes of the
mark. In both the strategies, Mahalanobis separation is utilized for the errand of arrangement. In [20], two
new strategies were displayed to remove the 1D signature highlights, in particular most extreme Length
Vertical Projection (MLVP) and Least Length Even Projection (MLHP). Both the strategies include in
turning the mark by an edge to make it revolution invariant. This was proficient by inspecting the edge in
the scope of [300, 300] and finding the point which relates to the most extreme (resp. least) length of the
vertical (resp. level) projection of the mark. The relating point is then used to adjust the mark. Next, the
vertical and even projections of the adjusted mark are treated as 1D feature, followed by afurther process
of size normalization using the length of the vertical projection. Signature coordinating and grouping is at
long last finished utilizing a broadened DTW procedure. Trial results, connected to their own datasets,
appeared fascinating results. For the above methodologies, the real confinements are taking after. Initially,
it is constrained to the little scope of mark revolution. Second, measure standardization procedure is
exceptionally delicate to clamor subsequent to the length of the vertical projection can be very diverse on
the off chance that some expansion objects (e.g., ink, blobs, and clamor) are nearness in the mark. At long
last, the utilization of vertical furthermore, flat projections as 1D elements is not adequately unmistakable
for segregating very comparable marks in which one of them could be made by an alternate author. Other
DTW-based methodologies that utilize the worldwide elements (e.g., Radon Change, Wavelet
examination) have been displayed. In synopsis, despite the fact that various systems have been completed
to manage the issue of disconnected from the net manually written mark cheque, it is untimely to infer that

the answer for this issue was finished. While signature highlight is a significant point prompting the
accomplishment of a disconnected from the net mark cheque, it appears that the leaving strategies utilized
entirely basic components. These strategies are not adequately particular, are touchy to scaling change, and
are incompletely invariant to turn change. Execution of these frameworks is exceptionally reliant on the
key stride of pre-preparing like commotion expulsion. Advance more, the existing techniques regularly
report the outcomes all alone datasets with a constrained correlation with different techniques. As the
objective of their trials is fundamentally worried with the sole motivation behind demonstrating the
exceptional aftereffects of the proposed technique, such an assessment convention is excessively subjective
and the genuine execution is not correctly assessed. Moreover, the majority of these techniques are keen on
settling on a simply double choice, which was known not a few disadvantages as said some time recently.
3. PROPOSED APPROACH
These certainties persuaded us to build up a disconnected from the net mark check framework whose
fundamental target is to give the client two yields: a proof score assessed from the info designs and the
level of assurance of finding that score. This kind of methodology alludes to the structure of settling on
delicate choice, as said in [16], and the framework is considered as a criminological penmanship master
that he looks at the info designs (e.g., commonly including one doubted signature and an arrangement of
real marks from a reference author) by means of two inverse speculations H0 and H1 expressed as takes
after. The proposed approach works with dark scale pictures. The info pictures are first experienced some
pre-handling steps including binarization, skeletonization, and testing. The Otsu strategy [12] is chosen for
the binarization venture as it is all around adjusted to the signature pictures which regularly contain two
predominant dimension levels relating to the mark content (i.e., frontal area) and the foundation.
Regardless, this binarization step will cause the loss of subtle element data. For instance, the dimension
levels in the mark pretty much uncover the weight and drive data of the composition procedure. These
lines misuse the skeleton-based elements for just the reason for extracting the worldwide components. The
subtle element data in the dark scale picture is still abused when figuring the nearby features. Starting from
the parallel mark picture, the skeleton of the mark is removed utilizing the strategy. This method permits
us to gauge the nearby line thickness at every skeleton point that we will utilize this quality as the scale
parameter for registering the neighbourhood highlights. The last pre-preparing step is testing. To pick up
time proficiency while applying the shape setting descriptor, the skeleton of a mark is consistently tested
yielding an arrangement of N test focuses. Every specimen point is then described by two parameters: the
area of the point and the nearby scale figured as the like thickness now. Just the specimen focuses are
considered to process the shape connection highlights and the nearby elements in the consequent stages.
4. METHODOLGY
The information acquistion is an imperative assignment in Signature Sample acknowledgment. With a
specific end goal to procure an adequate information we have been gathered the characters and filtered
utilizing scanner of determination 300 dpi. An aggregate of 1000 JPEG Signature Tests from 75 different
journalists of 100 words for each are acquired and put away in database for further examination.
4.1. Preprocessing
In pre-processing stages we did noise removal, binarization, edge detection and thinning operations are
performed on characters for better enhancement
4.2. Segmentation
The scanned handwritten Signature Samples are segmented into individual Signature Samples by using the
edge pixels technique and later each Signature Samples are combined into separate folders for separate
Signature Samples, where in each folder we have stored 100 samples for testing purpose totally 100 folders
of all Signature Samples.

Distance Classifier
5. FEATURE EXTRACTION
Feature extraction plays a vital role in improving the classification effectiveness and computational
efficiency. A. Edge Direction Distribution. The initial step, we distinguished the edge of the paired picture
utilizing Sobel Location method. These edge identified picture are checked utilizing 8 associated pixel
neighbourhood. After this the quantity of lines and segments in a paired picture is discovered utilizing size
function. Now the primary dark spot pixel in an picture is recognized and this pixel is considered as focus
pixel of the square neighbourhood. Next we checked the dark edge utilizing the coherent and administrator
as a part of all bearing beginning from the middle pixel and completion in any of the edge in square. To
maintain a strategic distance from the excess the upper two quadrants in the area is checked and it is hard
to distinguish the bearing of Signature Sample composed along the edge part. This will give the”n”
conceivable points. These confirmed edges of every pixel are numbered into n-paired histogram which is
then standardized to a likelihood circulation which gives the probability of an edge section situated in the
picture at the point measured from the level.
6. EDGE HINGE DISTRIBUTION
This system is helpful to distinguish the shape of the unmistakable styles from various journalists which is
ascertained with the assistance of neighbourhood edges along the edges. The Edge pivot feature considers
two edge pieces rising up out of focus pixel and join probability dispersion of introductions of two pieces
of a ”pivot” are ascertained. The Standardized histogram gives the joint likelihood dispersion for”pivoted”
edge pieces situated at the edges 1 and 2. The introduction is checked in 16 headings for a solitary edge.
Here we consider just the aggregate number of blends of two points of non-repetitive qualities and the
basic consummation pixels are killed.
7. FEATURE SELECTION
In this we made the classification based on Collective Characters Feature Selection. The agenda we took to
test, against zero, the difference µ1 and µ2 between the means of the values taken by a feature in two
classes. Let us take Class1 were
Table I Significance of Degrees of Freedom
X_ i, i=1,2,....,N, be the sample values of the feature in class ᵠ1 with µ1: Likewise, for Class2 were ᵠ2 we
shall take y i, i=1,2....,N with mean µ2: For our assumption the variance of the feature values is the same in
both classes σ2
_i= σ2
_i= σ2
For the purpose of closeness of two mean values, hypothesis test was carried
out
H_1: ∂µ=µ1-µ2#0
H_0: ∂µ=µ1-µ2=0
here x,y denote the random variables corresponding to the values of the feature in the two classes ᵠ1,ᵠ2
were statistical independence has been assumed. Likwise E[z]=µ1-µ2 and due to the independence

assumption σ2
–x=2σ2
We took the similar arguments were we used before, now we have this formula
given below.
and the known variance case z follows the normal
distribution for large N. In the table1 the values used to decide about the equation
If the variance is not known, then we choose the test static
Where
From this we can show that follows a chi-square distribution with 2N-2 degrees of
freedom.
7.1. Run Length Distribution
This run length circulation is utilized to decide on binarized pictures by thinking about of dark pixel which
is continuos on closer view or white pixel which is coordinating to foundation. The filtering methods are
two sorts: Flat & Vertical. In the Flat the lines of the picture and in the Vertical the section of the picture is
considered. Later the likelihood circulation is translated by utilizing the standardized histogram of run
lengths. The Orthogonal data to the directional components is acquired by utilizing the run lengths.
7.2. Length
The aggregate length of the every Signature Sample is distinguished in every section in the twofold picture
to locate the first and last pixels in the picture and store their segment number. At last discover the length
of the picture is computed by subtracting the section number of the last pixel to the segment number of the
primary pixel.
7.3. Height
The Tallness of the every Signature Sample is recognized continuously at every column in the parallel
picture. The first and the last pixels of the picture are found and the corresponding column numbers are put
away. At long last the tallness of the picture is computed by subtracting from the column number of last
pixel to the line number of first pixel.

Distance Classifier
1) Height from baseline to upper edge: By this system, we ascertained the tallness of the Signature
Sample from benchmark to upper edge is figured by deciding the gauge position of the Signature Sample.
This is finished by throwing an exhibit capacity, here the list is column number in the
Signature Sample. Next the quantity of dark pixels in every column is calculated and the outcomes are put
away in exhibit. At long last the whole Signature Sample, the most extreme estimation of the exhibit is
perceived and the corresponding grow number is put away as the basline. The length of the signature
sample from the benchmark to the upper edge is figured by subtracting the line number of first pixel in the
picture from its column number of the pattern.
2) Height from baseline to lower edge: By this strategy the tallness of the paired Signature Sample from
the pattern to the lower edge is controlled by ascertaining the gauge line number.
Next the line number of the last pixel of the Signature Sample is considered. The tallness of the picture
from the pattern to the lower edge is ascertained by subtracting the last pixel line number to the benchmark
line number.
8. CLASSIFIERS
For this experiment we used the Manhattan distance classifier & Euclidian Classifier. To calculate the
variation of different angles this classifiers is very powerful. For assumption Equi probable classes with the
same covariance matrix, g_x with the equation as shown below
where constants have been neglected. ∑=σ2
I : In this case maximum g i (x) implies minimum.
Euclidean distance: d _ε = || x_µi ||. Thus featue vectors are asigned to classes according to their Euclidean
distance from the respective mean points.
9. CONCLUSION AND FUTURE ENCHANCEMENT
The endeavor has been made to actualize Signature Test utilizing edge-based element that portrays the
adjustments in course attempted amid the written by hand Signature Tests by clients. It performs superior
to anything the various assessed highlights. Our test outcomes demonstrate that the best performing
highlights when a lot of content is accessible, to perform best contrasted with the others when little content
is accessible, not with standing or having significantly higher measurement. The outcomes reported here,
in view of a spotless information set, have a scholarly significance with regards to the handiness of various
components and they don’t completely address issues like size invariance, non-uniform foundation, and
debased records. The next future exploration of interest will be concentrating on decreasing the
dimensionality of the element vectors, and yet keeping their oppressive force on further enhancing
execution by consolidating diverse components with a specific end goal to abuse their in born level of
orthogonally.
10. RESULTS OBTAINED
Figure 3 Input with noise

Figure 4 Noise removed
Figure 5 Sobel operator specimen
Figure 6 Conture Speciman
Figure 7 X-stats
Figure 8 Imgoriginal datasets

Distance Classifier
Figure 9 Exact Matched Recognized ID Result
Figure 10 Path of the Exact Match Recognized ID Result opened in notepad
REFERENCES
[1] Belongie, S., Malik, J., Puzicha, J.: Shape matching and object recognition using shape contexts. IEEE
Trans. Pattern Anal. Mach. Intell.24(24), 509522 (2002).
[2] Blumenstein, M., Liu, X.Y., Verma, B.: A modified direction feature for cursive character
recognition.In: Proceedings of 2004 IEEE International Joint Conference on Neural Networks, vol. 4,
pp. 29832987 (2004)
[3] Brummer, N., Preez, J.d.: Application-independent evaluation of speaker detection. Computer Speech
and Language 20(2), 230275 (2006)
[4] Deng, P.S., Liao, H.-Y.M., Ho, C.W., Tyan, H.-R.: Wavelet-based off-line handwritten signature
verification. Comput. Vis. Image Underst. 76(3),173190 (1999)
[5] di Baja, G.S.: Well-shaped, stable, and reversible skeletons from the (3,4)-distance transform. J. Visual
Comm. and Image Representation 5(1), 107115 (1994)
[6] El-Yacoubi, A., El Yacoubi, A., Bortolozzi, F., Sabourin, R.: An offline signature verification system
using hidden markov model and cross-validation. In: Proceedings of the 13th Brazilian Symposium on
Computer Graphics and Image Processing, pp. 105112 (2000).
[7] Fang, B., Leung, C.H., Tang, Y.Y., Kwok, P.C.K., Tse, K.W., Wong, Y.K.: Offline signature
verification with generated training samples. IEE Proceedings of Vision, Image and Signal Processing
149(2), 8590 (2002).
[8] Fang, B., Leung, C.H., Tang, Y.Y., Tse, K.W., Kwok, P.C.K., Wong, Y.K.: Offline signature
verification by the tracking of featute and stroke positions. Pattern Recog. 36(1), 91101 (2003)
[9] Fang, B.,Wang, Y.Y., Leung, C.H., TSE, K.W., Tang, Y.Y., Kwok, P.C.K.,Wong, Y.K.: Offline
signature verification by the analysis of cursive strokes. Int. J. Pattern Recognit. Artif. Intell. 15(4),
659673 (2001)
[10] Fischler, M.A., Bolles, R.C.: Random sample consensus: a paradigm for model fitting with applications
to image analysis and automated cartography. Communication ACM 24(6), 381395 (1981)
[11] Gilperez, A., Alonso-Fernandez, F., Pecharroman, S., Fierrez, J., Ortega-Garcia, J.: Off-line signature
verification using contour features. In: Proceedings of the 11th International Conference on Frontiers in
Handwriting Recognition (2008)

[12] Gonzalez-Rodriguez, J., Fierrez-Aguilar, J., Ramos-Castro, D., OrtegaGarcia, J.: Bayesian analysis of
fingerprint, face and signature evidences with automatic biometric systems. Forensic Sci. Int. 155(2),
126140 (2005)
[13] Impedovo, D., Pirlo, G.: Automatic signature verification: The state of the art. IEEE Transactions on
Systems, Man, and Cybernetics, Part C: Applications and Reviews 38(5), 609635 (2008)
[14] Jena, D., Majhi, B., Panigrahy, S.K., Jena, S.K.: Improved offline signature verification scheme using
feature point extraction method.In: Proceedings of the 7th IEEE International Conference on Cognitive
Informatics, pp. 475480 (2008)
[15] Jonker, R., Volgenant, A.: A shortest augmenting path algorithm for dense and sparse linear assignment
problems. Computing 38(4), 325340(1987)
[16] Dr. Vangala Padmaja, A Brief Review on Hand Written Character Recognition, International Journal of
Advanced Research in Engineering and Technology (IJARET), 5(2), 2014, pp. 70–78.
[17] Manju and Abhishek Bhatnagar. Application of Speech with Their Analysis about Recognition.
International Journal of Computer Engineering and Technology (IJCET), 6(10), 2015, pp. 66-80.

OFFLINE HANDWRITTEN SIGNATURE RECOGNITION USING EDGE HINGE AND EDGE EXTRACTION TECHNIQUES AND MANHATTAN DISTANCE CLASSIFIER

Recommended

Recommended

More Related Content

Viewers also liked

Viewers also liked (20)

Similar to OFFLINE HANDWRITTEN SIGNATURE RECOGNITION USING EDGE HINGE AND EDGE EXTRACTION TECHNIQUES AND MANHATTAN DISTANCE CLASSIFIER

Similar to OFFLINE HANDWRITTEN SIGNATURE RECOGNITION USING EDGE HINGE AND EDGE EXTRACTION TECHNIQUES AND MANHATTAN DISTANCE CLASSIFIER (20)

More from IAEME Publication

More from IAEME Publication (20)

Recently uploaded

Recently uploaded (20)

OFFLINE HANDWRITTEN SIGNATURE RECOGNITION USING EDGE HINGE AND EDGE EXTRACTION TECHNIQUES AND MANHATTAN DISTANCE CLASSIFIER