8/3/2019 5 Keerthi Report New
1/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 1
CHAPTER-1
INTRODUCTION
Since the inception of computers we are witnessing a great deal of research activities
in the field of computer human interface. The input devices such as keyboard and
mouse have limitations in comparison with input through natural handwriting. The
natural handwriting is a very easy way of exchanging information between computers
and human beings. Also, it is difficult to input data to computers for scripts Chinese
and Japanese as these scripts have a large number of alphabets. It is also difficult to
input data for computers for Indian language scripts owing to their complex typing
nature. Two quick and natural ways of communication between users and computers
are inputting the data through handwritten documents and through speech. Speech
recognition has limitations in noisy environment and especially where privacy of an
individual is required. This work focuses on the problem of handwriting recognition
only.
1.1 Handwriting recognition
The technique by which a computer system can recognize characters and other
symbols written by hand in natural handwriting is called handwriting recognition
system [1]. Handwriting recognition has been a popular area of research since few
decades under the purview of pattern recognition and image processing. Handwriting
recognition can be broken down into two categories: off-line and on-line.
Off-line
Off-line character recognition takes a raster image from a scanner (scanned
images of the paper documents), digital camera or other digital input sources. The
image is binarized through threshold technique based on the color pattern (color or
gray scale), so that the image pixels are either 1 or 0 [2].
On-line
In on-line, the current information is presented to the system and recognition
(of character or word) is carried out at the same time. Basically, it accepts a string of
( x, y) coordinate pairs from an electronic pen touching a pressure sensitive digital
tablet.
8/3/2019 5 Keerthi Report New
2/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 2
On-line handwriting recognition system, by contrast, captures the temporal or
dynamic information of the writing, enhances the accuracy over off-line. Another
advantage is interactivity, which means recognition errors can be corrected
immediately with the series of test. Yet, adaptation of any drawings of character is
also an advantage over off-line. When the user faces that some characters are notrecognized accurately, user can alter his way of drawing until it recognizes. It means
user can adapt to the machine. Conversely, recognizers are capable of adapting users
drawing, usually by storing possible samples from a large number of users for
subsequent recognition. Thus, there is adaptation of user to machine and of machine
to user. Electronic pen input is the direct method to compare with the both off-line
and key-board entry to the system having recognition intelligence. In addition, on-line
recognition improves the work-flow, the information is immediately available.
However, the natural and comfortable style in writing effectively reduces difficulty atthe threshold of using computers for common users. Moreover, it is recently showed
that handwriting input is the most acceptable and welcomed input style.
A handwriting recognition system can further be broken down into two
categories of writer independent and writer dependent.
Writer Independent and Writer Dependent
A writer independent recognition system recognizes wide ranges of possible writing
styles, while a writer dependent recognition system is trained to recognize only from
specific users. Therefore, a writer dependent recognition system works on data with a
smaller variability and therefore a chance of having higher reliability is achieved in
contrast to writer independent recognition system. Writing ones style brings
unevenness in writing units, which is the most difficult part. Variability in stroke
numbers, their order, shape and size, tilting angle and similarity among characters
from one another are found more often in writer independent recognition system.
Broadly, there are two kinds of writing styles. They are hand printed and
cursive handwriting. In cursive style, strokes are deliberately linked forming one from
many to draw the character, while in hand printed style possible number of strokes are
used, each stroke has significant role to complete the character. In cursive style, the
important information such as intersections, loops, curves, straight lines and hooks
etc. are missing. Sometimes, both writing styles are mixed. Natural handwriting
contains all types of styles in writing from any of the users. Specifically, the writing is
said to be natural as if users write on a piece of paper.
8/3/2019 5 Keerthi Report New
3/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 3
With the introduction of portable hand held computers and computing devices
such as PDAs (Personal Digital Assistant), non-keyboards and non-keypads based
methods for inputting data are receiving more interest in both academic and
commercial research communities. The most promising options are pen based and
voice based inputs. Pen based method in inputting can be either off-line or on-line.
1.2 Challenges in online handwriting recognition system
A variation in handwriting is the prominent problem and achieving high degree of
accuracy is a tedious task. These variations are caused by different writing styles.
Variation in handwriting among different writers occurs since each writer possesses
own speed of writing, different styles, sizes or positions for characters or text.
Variation in handwriting styles also exists within individual persons handwriting.
This variation may take place due to: writing in various situations that may or may not
be comfortable to writer; different moods of writer; style of writing same characters
with different shapes in different situations or as a part of different words; using
different kinds of hardware for handwriting.
1.2.1 Handwriting styles variations
Handwriting styles variations depend on alignments and the different form of
characters. These variations are geometrical in nature. Common geometrical
properties are position, size, aspect ratio of strokes or characters, retraces, slant of
strokes and number of strokes in a character. Fig. 1.1 illustrates the few samples of
Gurmukhi characters from five different writers. One can note that variations exist in
each sample of a character. Fig. 1.2 illustrates five samples of few characters of
Gurmukhi script from individual writer. One can note that some kind of variations
also exists in each sample of a character although such samples share high degree of
similarities. The shape of a character is also influenced by the word in which it is
appearing. Characters can look similar although their number of strokes, and the
drawing order and direction of the strokes may vary considerably [3].
1.2.2 Constrained and unconstrained handwriting
Handwriting styles could be constrained or unconstrained [4]. Constrained
handwriting is boxed discrete and spaced discrete in nature. Unconstrained
handwriting is cursive or mixed cursive in nature. In boxed discrete handwriting, each
character is written inside a special box. Fig. 1.3 illustrates the boxed discrete
8/3/2019 5 Keerthi Report New
4/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 4
handwriting. When each character is written separately with spaces and no character
touches other character is called spaced discrete handwriting. If each character is
written separately and touches other characters, it is referred as run-on discrete
handwriting. When characters in one word are connected and strokes are used more
than once in individual character, it is referred to cursive handwriting.It is observed that most of the people write in mixed cursive styles that
includes mixture of spaced, run-on discrete and cursive styles handwriting. Spaced
discrete, run-on discrete, cursive and mixed cursive handwriting styles are illustrated
in Fig. 1.4.
It is a difficult task to recognize cursive handwriting due to great amount of
variability. Each writer is having ones own speed of writing and uses different shapes
to represent characters. Also, in cursive handwriting no clear boundaries are specified
between characters to distinguish between them.
Fig. 1.1: Variation in characters handwritten by five writers.
8/3/2019 5 Keerthi Report New
5/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 5
Fig. 1.2: Variation in characters handwritten by same writer.
Fig. 1.3: Boxed discrete handwriting.
Fig. 1.4: Different styles of handwriting.
8/3/2019 5 Keerthi Report New
6/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 6
1.2.3 Personal, situational and material factors
The personal factors in handwriting variations include writers handedness. A writer
is either left handed or right handed. It has been noted that left and right handed
people use different positions and directions in handwriting. A good recognition
requires neat and clean handwriting. In most of the cases, it has been noted that neatand clean handwriting does not take place as handwriting of people also depends on
their profession [5].
The situational factors depend on the way of presentation of writing. The way
of presentation could be stressful or in haste or distraction while writing. The material
factors depend on the hardware used in writing. The material used in writing may
provide comfort or discomfort to writer that result into variations in handwriting. This
includes the position and size of writing board. The length of the writing line or the
size of the writing boxes for characters could have effect on the handwriting style.
1.3 Applications
Character and handwriting recognition has a great potential in data and word
processing for instance, automated postal address and ZIP code reading, data
acquisition in bank checks, processing of archived institutional records, etc.
Combined with a speech synthesizer, it can be used as an aid for people who are
visually handicapped. As a result of intensive research and development efforts,
systems are available for English language. However, less attention had been given to
Indian language recognition. Main reasons for this slow development could be
attributed to the complexity of the shape of Indian scripts, and also the large set of
different patterns that exist in these languages, as opposed to English.
1.4 Problem Statement
This work aims at developing a real-time on-line handwriting recognition system for
Indian languages - Hindi on handheld device such as mobile phone. The system has to
recognize all characters of Hindi alphabet and display the recognized character in
Hindi font in the device display area.
8/3/2019 5 Keerthi Report New
7/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 7
1.5 Objectives
Objectives of the system are as below:
To develop a system that will correctly and efficiently recognizes and displaythe handwritten characters on display area of the handheld device.
System should be writer independent. System should be memory and processing efficient.
1.6 Expected Outcome
The system developed should be able to capture handwritten characters of the users
drawn in the handheld device. The characters have to be recognized within small
duration of time and displayed in Hindi font in the display area of the handheld
device.
8/3/2019 5 Keerthi Report New
8/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 8
CHAPTER-2
SYSTEM METHODOLOGY
The established procedure to recognize online handwritten characters includes
following phases: data collection, preprocessing, feature extraction or computation of
features, segmentation, recognition and post-processing. This Chapter discusses each
phase used in a typical on-line handwriting recognition system. The output obtained
from one phase becomes input for the next phase. These phases are illustrated in
Fig. 2.1.
Fig. 2.1: Phases of handwriting recognition.
2.1 Data collection
Online handwriting recognition requires a transducer that captures the writing as it is
written. The most common of these devices is the electronic tablet or digitizer. These
Data Collection
Preprocessing
Feature Extraction
Segmentation
Recognition
Post - Processing
8/3/2019 5 Keerthi Report New
9/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 9
devices uses a pen that is digital in nature. Data collection is the first phase in online
handwriting recognition that collects the sequence of coordinate points of the moving
pen. A typical pen includes two actions, namely, PenDown and PenUp. The
connected parts of the pen trace between PenDown and PenUp is called a stroke.
These pen traces are sampled at constant rate, therefore these pen traces are evenlydistributed in time and not in space. The common names of electronic tablet or
digitizer are personal digital assistant, cross pad and tablet PC. The appearances of
personal digital assistant, cross pad and tablet PC are shown in Fig. 2.2.
Fig. 2.2: Commonly used hardware devices for capturing handwriting.
The selection of these hardware devices is mainly based on compatibility with
operating system in use, active area dimensions and report rate of pen movements.
2.2 Preprocessing
Preprocessing phase in handwriting recognition is applied to remove noise or
distortions present in input text due to hardware and software limitations in
comparison withsmooth handwriting. These noise or distortions include irregular size
of text, missing points during pen movement collections, jitter present in text, left orright bend in handwriting and uneven distances of points from neighboring positions.
In online handwriting recognition, preprocessing includes five common steps,
namely, size normalization and centering, interpolating missing points, smoothing,
slant correction and re-sampling of points. These steps are illustrated in Fig. 2.3.
8/3/2019 5 Keerthi Report New
10/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 10
Fig. 2.3: Common steps in preprocessing phase.
2.2.1 Size normalization and Centering
Size of the input stroke depends on how user moves the pen on writing pad. Stroke is
not generally centered when the pen is moved along the border of writing pad. Size
normalization and centering of stroke is a necessary process that should be performed
in order to recognize a character [6].
This can be achieved by comparing input stroke border frame with assumed
fixed size frame and further can be moved along with the assumed center location.
Fig. 2.4 illustrates size normalization.
Preprocessing Phase
Size normalization and
centering
Interpolating missing
points
Smoothing
Slant correction
Resampling of points
8/3/2019 5 Keerthi Report New
11/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 11
Fig. 2.4: (a & c) Input character (b & d) size normalization and centering.
2.2.2 Interpolating missing points
The stroke drawn with high speed will have missing points. These missing points can
be calculated using various techniques such as Bezier and B - Spline [7]. In piecewise
interpolation technique, a set of consecutive four points is considered for obtaining
the Bezier curve. The next set of four points gives the next Bezier curve. Fig.2.5
illustrates this step.
Fig. 2.5: Interpolating missing points.
8/3/2019 5 Keerthi Report New
12/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 12
2.2.3 Smoothing
Flickers exist in handwriting because of individual handwriting style and the
hardware used. These flickers can be removed by modifying each point of the list
with mean value of k-neighbors and the angle subtended at position from each end
[8].
2.2.4 Slant correction
Handwritten words are usually slanted or italicized due to the mechanism of
handwriting and the personality. The main techniques for slant estimation and
correction are run length based technique, projection method, extrema method and
generalized chain code estimator [9]-[12]. Fig.2.6 illustrates smoothing and slant
correction.
Fig. 2.6: Stroke before and after slant correction.
2.2.5 Resampling of points
Resampling of points is required to keep the points in the list at equal distances, as far
as possible. For any pair of points in the list having a distance greater than one, we
add a new point between such pairs. Any pairs having distance less than one is
untouched. The list obtained after the resampling of points is preprocessed. Fig. 2.7
shows the shape of stroke after applying the process of resampling of points [13].
8/3/2019 5 Keerthi Report New
13/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 13
Fig. 2.7: Stroke before and after resampling of points.
2.3 Feature extraction
In the process of handwriting recognition, it is important to identify correct features.
Feature extraction is essential for efficient data representation and for further
processing [14]. Also, high recognition performance could be achieved by selecting
suitable feature extraction method. Computational complexity of a classification
problem can also be reduced if suitable features are selected.
Features vary from one script to another script and the method that gives better
result for a particular script cannot be applied for other scripts. Also, there is no
standard method for computing features of a script.
In feature extraction stage each character is represented as a feature vector,
which becomes its identity. The major goal of feature extraction is to extract a set of
features, which maximizes the recognition rate with the least amount of elements. Due
to the nature of handwriting with its high degree of variability and imprecisionobtaining these features, is a difficult task. Features are classified into two categories,
namely, low-level and high-level features also called Statistical and Structural
features respectively.
8/3/2019 5 Keerthi Report New
14/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 14
2.3.1 Structural features
Structural features are those which provide useful information such as loops,
crossings, headline, straight line and dots. These features are derived on the basis of
calculating low-level features such as directions, positions, slope, area and slant etc. in
a stroke.
Loops
Loop recognition includes three stages: first stage finds the existence of a loop in an
alphabet, second stage determines direction (left or right), third stage finds position
(top, mid or bottom) with respect to upper-middle and middle-lower zones partitions.
The existence of a loop in an input handwritten stroke is shown in Fig. 2.4.
Overwriting in a single stroke can also result into a loop, preprocessing removes such
redundancy in data and also such loops can be avoided with the knowledge of loop
position, loop area, loop width and loop height.
Crossings
Crossings, as shown in Fig. 2.8, are the intersection of strokes. Number of crossings
in a character becomes meaningful when two or more strokes intersect. Position for
the point of intersection classifies characters into various subgroups that are further
helpful in narrowing the search operation.
Headline
Headline exists at the partition of upper-middle region and is horizontal in nature as
shown in Fig. 2.8. The identification of headline stroke narrows the search operation
as character is considered as a group of strokes. Horizontal nature of headline stroke
is computed on the basis of direction, curliness, position, slope and non-loop nature of
stroke.
Straight line
Straight line, as shown in Fig. 2.8, exists in some of the characters of Hindi script.
The straight line can be found on the basis of direction, position, crossings, headline,
curliness and non-loop nature of a stroke.
Dots
Dots are the isolated strokes as illustrated in Fig. 2.9. The dot feature can be identified
on the basis of stroke length, stroke direction, stroke position and nature of points in
strokes.
8/3/2019 5 Keerthi Report New
15/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 15
Fig. 2.8: Positions of headline, crossing and straight line in a character.
Fig. 2.9: Characters with dot feature.
2.3.2 Statistical features
The followings are the major statistical features used for character representation:
Zoning
The frame containing the character is divided into several overlapping or non-
overlapping zones. The densities of the points or some features in different regions are
analyzed and form the representation. For example, contour direction features
measure the direction of the contour of the character, which are generated by dividing
the image array into rectangular and diagonal zones and computing histograms of
chain codes in these zones. Another example is the bending point features, which
represent high curvature points, terminal points and fork points. Fig. 2.10 indicates
contour direction and bending point features.
8/3/2019 5 Keerthi Report New
16/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 16
Fig. 2.10: Contour direction and bending point features with zoning.
Crossings and Distances
A popular statistical feature is the number of crossing of a contour by a line segment
in a specified direction. One of the method can be, the character frame is partitioned
into a set of regions in various directions and then the black runs in each region are
coded by the powers of two. Another method can be to encode the location and
number of transitions from background to foreground pixels along vertical lines
through the word. Also, the distance of line segments from a given boundary, such as
the upper and lower portion of the frame, can be used as statistical features. These
features imply that a horizontal threshold is established above, below and through the
center of the normalized script. The number of times the script crosses a threshold
becomes the value of that feature. The obvious intent is to catch the ascending and
descending portions of the script.
Projections
Characters can be represented by projecting the pixel gray values onto lines in various
directions. This representation creates one-dimensional signal from a two dimensional
image, which can be used to represent the character image.
The next type of the statistical featuring is that the profiles. Profiles means the
distance between the bounding box of the character image and the edge of the
character. This distance is responsible for the proper reconstruction of the character in
the text format. The letters can be distinguished only if the profiles are there. Consider
that the letters p and q. Profiles are used for the detection of the contour of the
character image. The profiles distance gives the uppermost and the lowermost points
of the contour. It also helpful in maintaining the structure of the character. These are
all about the statistics of the feature extraction. The Fig. 2.11 shows projection and
profiles.
8/3/2019 5 Keerthi Report New
17/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 17
Fig. 2.11: Projection and Profiles.
2.4 Segmentation
Segmentation is one of the phases of handwriting recognition in which data are
represented at character or stroke level so that nature of each character or stroke can
be studied individually. Segmentation is classified into two categories: external
segmentation and internal segmentation. External segmentation is performed prior to
recognition. Segmentation performed during the process of recognition is called
internal segmentation. External segmentation provides greater interactivity, savings of
computation, and simplifies the job of the recognizer. Plamondon and Srihari [15]
presented a survey on handwriting recognition systems where segmentation has been
discussed for both offline and online handwriting recognition systems.
It has been noted that segmentation study in offline handwriting recognition
system is beneficial to understand segmentation in online handwriting recognition
system as word level segmentation is one of the common task in offline and online
handwriting recognition systems. Both offline and online handwriting recognition
systems identify characters or strokes in word level segmentation.The procedure to
segment a Hindi word into characters (including core characters, and top and bottom
modifiers) is illustrated in Fig.2.12.
8/3/2019 5 Keerthi Report New
18/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 18
Fig. 2.12: Procedure of Hindi character recognition.
The numbered arrow in Fig.2.12 represents the step of segmentation, and the
characters with solid bounding boxes are the final segmentation results. The
procedure to do character segmentation can be described as follows:
Step 1: Locate the header line and separate the core-bottom strip which contains the
core strip and bottom strip, and a top strip which contains the header line and top
modifiers.
Step 2: Identify core strip and bottom strip from the core-bottom strip, and extract
low modifiers.
Step 3: Separate core strip into characters which may contain conjunct/shadow
characters.
Step 4: Segment conjunct/shadow characters into single characters.
Step 5: Remove the header line from the top strip and extract top modifiers.
Step 6: Put header line back to the segmented core characters.
8/3/2019 5 Keerthi Report New
19/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 19
2.5 Recognition
Statistical, syntactical and structural, neural network and elastic matching are the
commonly used handwriting recognition methods [16]. They are explained in the
following subsections.
2.5.1 Statistical methods
In statistical approach, each pattern is represented in terms of features and is viewed
as a point in dimensional space. This involves selection of features that all pattern
vectors belonging to different categories or classes to occupy disjoint region in a
dimensional space. The statistical methods are based on prior probabilities of classes
and assume variations in natural handwriting as stochastic in nature. Statistical
methods are classified as parametric and non-parametric methods. In parametric
methods, handwriting samples are statistical variables from distribution that is
characterized by a set of parameters and each class includes its own distribution
parameters. The selection of parameters is based on training data. Hidden Markov
Model (HMM) is the common example of parametric methods. Non-parametric
methods are directly estimated from training data. The nearest neighbors are the
common non-parametric methods.
Parametric methods are preferred as compared to non-parametric methods as
parametric methods are computationally easier than non-parametric methods. HMM is
the most widely used parametric statistical method applied to online handwriting
recognition systems. Initially, HMMs were applied to speech recognition. The HMMs
became popular in online handwriting recognition systems in early 1990s. HMMs
found to be suitable for cursive handwriting. The results obtained using HMMs are
reliable as outcomes are numerical values and there is always a scope to improve
recognition system using HMMs.
2.5.2 Structural and syntactical methods
Structural and syntactical methods are related to handwritten patterns where structures
and grammar are considered. Structural recognition provides a description of how the
given pattern is constructed from the primitives. This paradigm has been used in
situations where the patterns have a definite structure which can be captured in terms
of a set of rules, such as waveforms, textured images, and shape analysis of contours.
In syntactic pattern recognition, a formal analogy is drawn between the structure of
patterns and the syntax of a language. The patterns are viewed as sentences belonging
8/3/2019 5 Keerthi Report New
20/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 20
to a language, primitives are viewed as the alphabet of the language, and the sentences
are generated according to a grammar. Thus, a large collection of complex patterns
can be described by a small number of primitives and grammatical rules. The
grammar for each pattern class must be inferred from the available training samples.
The chain codes are widely used structural representations of online handwriting.Chain code means that a stroke is temporarily divided into segments and segments are
coded. These segments are small straight lines of equal lengths and consider
information as directions, angles, geometric information in segments.
2.5.3 Neural network methods
Neural networks can be viewed as parallel computing systems consisting of an
extremely large number of simple processors with many interconnections. Neural
network models attempt to use some organizational principles such as learning,
generalization, adaptability, fault tolerance, distributed representation and
computation in a network of weighted directed graphs in which the nodes are artificial
neurons and directed edges are connections between neuron outputs and neuron inputs.
The main characteristics of neural networks are that they have the ability to learn
complex nonlinear input-output relationships, use sequential training procedures and
adapt themselves to the data. The most commonly used neural networks for pattern
classification tasks are feed-forward network, which include multilayer perceptron
and radial basis function networks. These networks are organized into layers and have
unidirectional connections between the layers. Another popular network is the self-
organizing map or kohonen network.
2.5.4 Elastic matching methods
Elastic matching is a generic operation in pattern recognition which is used to
determine the similarity between two entities. The pattern to be recognized is matched
against the stored template while taking into account all allowable changes. The
similarity measure can be optimized based on available training set. Elastic matching
is feasible but with availability of faster processors. Elastic matching is often called
deformable template, flexible matching, or nonlinear template matching [17]. Elastic
matching works very well for writer dependent data and does not require a relatively
large amount of training data [18].
8/3/2019 5 Keerthi Report New
21/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 21
2.6 Post-processing
Post-processing refers to the procedure of correcting misclassified results by applying
linguistic knowledge. All the possible outcomes of an individual character are studied
in terms of graph and the best suitable nature of character is depicted. Post-processing
is processing of the output from shape recognition. Language information can
increase the accuracy obtained by pure shape recognition. For handwriting input,
some shape recognizers yield a single string of characters, while others yield a
number of alternatives for each character, often with a measure of confidence for each
alternative. A postprocessor can operate on this information to obtain estimates for
larger linguistic units, such as words. When the shape recognizer yields a single
choice for each character, string correction algorithms are applicable. Alternate
choices provide more information for post-processing. In post-processing, a
dictionary can be used to restrict the character combinations. This can be
implemented as a grammar that specifies all possible combinations of characters.
8/3/2019 5 Keerthi Report New
22/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 22
CHAPTER-3
LITERATURE SURVEY
This chapter provides various schemes that are reported in the recent works for online
handwriting recognition. In the following section, we have summarized the schemes
mainly under two categories such as feature based and classifier based.
3.1 Classifier based schemes
Different classifiers such as DTW (Dynamic time warping), SVM (Support Vector
Machine), HMM, Neural network, etc., have been used for Online handwriting
recognition. The work done using these approaches is reviewed below.
3.1.1 DTW classifier
Dynamic Time Warping (DTW) classifier that can be used for handwriting
recognition is able to compare two curves in a way that makes sense to humans. As
can be seen in Fig. 3.1, DTW can compare characters in a way that is similar to the
way humans compare characters.
Fig. 3.1: Comparison of two curves using one to one comparison and DTW.
In the recognition system reported by Muralikrishna Sridhar, Dinesh
Mandalapu & Mehul Patel [19], a classifier called Active-DTW has been proposed. In
the experiments they used the database, which contains both online and off-line
handwriting information. They obtained an accuracy of 97.1% for Digits, 86.9% for
Lower case and 94.1% for Upper case English letters. Their system combines the
advantages of generative and discriminative classifiers to address the similarity of
between-class samples, while taking into account the variability of writing styles
within the same character class.
However, in order to create accurate models, a large number of training
samples is needed up front, which is not desirable or available in many practical
8/3/2019 5 Keerthi Report New
23/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 23
applications. Vandana Roy et. al., [20] proposed a supervised adaptation framework
for the Active-DTW classifier which allows recognition to begin with a small number
of training samples, and adapts the classifier to the new samples presented to the
system during recognition. They compare the performance of Active-DTW using the
proposed adaptation framework, with a nearest-neighbor classifier, on the onlinehandwritten Tamil character dataset.
A novel framework to adapt the Active-DTW classifier has been proposed by
Vandana Roy et. al. [20]. Such an adaptation framework is particularly useful for
rapidly deploying recognizers for new scripts, since the initial requirements for
training data can be very small. A considerable improvement in accuracy of the
Active-DTW over time was shown empirically using the proposed adaptation
strategy. Also, the performance was shown to be comparable with a nearest neighbor
classifier. The memory requirements for the model data file for Active-DTW
recognizer and time taken for adaptation was shown to be significantly less as
compared to the nearest neighbor classifier.
DTW classifier reported by Muralikrishna Sridhar et. al., is for English
characters, so to enhance the recognition system for Indian script Niranjan et. al., [21]
proposed a recognition system for Tamil handwritten text. In their system they used
subspace and DTW classifiers so as to combine the advantages of the two schemes to
formulate a hybrid scheme for recognition and carried out the experiment on all the
three modes, namely, writer dependent (WD), writer independent (WI) and writer
adaptive (WA). The system proposed is not suitable for commercial use because the
objective is only to reap the advantages of both the methods.
DTW-implementation that is suitable for the automatic recognition of Tamil
handwriting is proposed by Ralph Niels and Louis Vuurpij [22] in which a prototype
based classifier that uses DTW both for generating prototypes and for calculating a
list of nearest prototypes is used. Prototypes were automatically generated and
selected. The recognition system listed above consume more space and processing
time, so to reduce both Muzaffar Bashir, and Jurgen Kempf [23] proposed a system
which uses a Dynamic Time Warping technique which reduces significantly the data
processing time and memory size of multi-dimensional time series sampled by the
biometric smart pen device BiSP. It is found that the performance of the RDTW
(Reduced DTW) method complies very well with the claims of an online recognition
system. Single characters and PIN words, handwritten by the same person can be
8/3/2019 5 Keerthi Report New
24/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 24
recognized at an extremely high score (better 99%) with a responds time of less than
0.5 seconds. These excellent results lead to a promising application, namely the
biometric recognition of PIN codes being an essential part of the biometric two factor
person authentication method, where biometric person and PIN code recognition is
combined.
3.1.2 SVM classifier
SVM in its basic form implement two class classifications. It has been used in recent
years as an alternative to popular methods such as neural network. The advantage of
SVM, is that it takes into account both experimental data and structural behavior for
better generalization capability based on the principle of structural risk minimization
(SRM). The principle of an SVM is to map the input data onto a higher dimensional
feature space nonlinearly related to the input space and determine a separating hyper
plane with maximum margin between the two classes in the feature space.
The recognition system reported by Abdul Rahim Ahmad et. al., [24] uses
SVM for online handwriting recognition for English characters. In Online
Handwritten Character Recognition system for Devanagari and Telugu Characters,
proposed by H. Swethalakshmi et. al., [25] Support vector machines have been used
for constructing the stroke recognition engine. The main disadvantage of their
approach is that the set of baseline strokes that loops have been found to give the
lowest recognition accuracy. Swethalakshmi et. al., explored three approaches for
stroke recognition.
Single Recognition Engine approach in which each stroke is represented as an
n-dimensional feature vector depending on the choice of the number of points for
stroke representation. The features chosen to represent the curve are the co-ordinates
of points in the preprocessed stroke. Multiple SVM Engines approach in which
strokes are pre-classified into two categories based on a threshold set for curve length.
Further an SVM-based engine is constructed for each stroke category. Strokes with
curve length below the threshold are classified as small strokes. Small strokes are
subjected to normalization and smoothing Third approach is Stroke Recognition using
HMMs in which strokes are represented as variable-length sequences of frames. Each
frame consists of a feature vector representing the features captured at the
corresponding time instant. Here, they have used the co- ordinates of the point as
features for a frame.
8/3/2019 5 Keerthi Report New
25/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 25
Recognition systems listed above do not recognize mathematical symbols; to
support this Birendra Keshari and Stephen M. Watt [26] proposed Online
Mathematical Symbol Recognition using SVMs with Features from Functional
Approximation. The experimental results show that the SVM trained using features
from functional approximation produces results comparable to the other SVM basedrecognition system. This makes the functional approximation technique interesting
and competitive since the features have certain computational advantages.
3.1.3 HMM classifier
A hidden Markov model (HMM) is a statistical Markov model in which the system
being modeled is assumed to be a Markov process with unobserved (hidden) states.
An HMM can be considered as the simplest dynamic Bayesian network. In a regular
Markov model, the state is directly visible to the observer, and therefore the state
transition probabilities are the only parameters. In a hidden Markov model, the state is
not directly visible, but output, dependent on the state, is visible. Each state has a
probability distribution over the possible output tokens. Therefore the sequence of
tokens generated by an HMM gives some information about the sequence of states. A
hidden Markov model can be considered a generalization of a mixture model where
the hidden variables (or latent variables), which control the mixture component to be
selected for each observation, are related through a Markov process rather than
independent of each other.
Hidden Markov Models (HMM) have long been a popular choice for Western
cursive handwriting recognition following their success in speech recognition. Even
for the recognition of Oriental scripts such as Chinese, Japanese and Korean, Hidden
Markov Models are increasingly being used to model substrokes of characters.
However, when it comes to Indic script recognition, the published work employing
HMMs is limited, and generally focused on isolated character recognition. In this
effort, a data-driven HMM-based online handwritten word recognition system for
Tamil, an Indic script, is proposed by Bharath A and Sriganesh Madhvanath [27]. The
accuracies obtained ranged from 98% to 92.2% with different lexicon sizes. These
initial results are promising and warrant further research in this direction. The results
are also encouraging to explore possibilities for adopting the approach to other Indic
scripts as well. So to support another popular Indian language namely Telugu
Jagadeesh Babu et. al., [28] proposed a recognition system which is based on HMM
and uses a combination of time-domain and frequency-domain features. The system
8/3/2019 5 Keerthi Report New
26/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 26
gives top-1 accuracy of 91.6% and top-5 accuracy of 98.7% on a dataset containing
29,158 train samples and 9,235 test samples.
Parui et. al. [29] uses HMM to support Online Handwritten Bangla Character.
They used a database of 24,500 online handwritten isolated character samples written
by 70 persons. Samples in this database are composed of one or more strokes and wehave collected all the strokes obtained from the training samples of the 50 character
classes. These strokes are manually grouped into 54 classes based on the shape
similarity of the graphemes that constitute the ideal character shapes. Strokes are
recognized by using hidden Markov models (HMM). One HMM is constructed for
each stroke class. A second stage of classification is used for recognition of characters
using stroke classification results along with 50 lookup-tables (for 50 character
classes).
3.1.4 Neural Network classifier
Neural Nets (NN) and Hidden Markov Models (HMM) are the popular, amongst the
techniques which have been investigated for handwriting recognition. It has been
observed that NNs in general obtained best results than HMMs, when a similar feature
set is applied. The most widely studied and used neural network is the Multi-Layer
Perceptron (MLP). Such an architecture trained with back-propagation is among the
most popular and versatile forms of neural network classifiers and is also among the
most frequently used traditional classifiers for handwriting recognition.
Other architectures include Convolutional Network (CN), Self-Organized
Maps (SOM), Radial Basis Function (RBF), Space Displacement Neural Network
(SDNN), Time Delay Neural Network (TDNN), Quantum Neural Network (QNN),
and Hopfield Neural Network (HNN). Few attempts have been found in the literature
in which counter-propagation (CPN) architecture has been used for the recognition of
handwritten characters. Ahmed et. al., [30] made an attempt but only for digit
recognition.
Muhammad Faisal Zafar et. al., [31] proposed On-line Handwritten Character
Recognition system for upper case English alphabets which is an implementation of
Counter propagation Neural Net. CPN is more economical than convergence of other
NN architectures e.g. back-propagation where the training time can take long time.
The experiments provided the authors an opportunity to explore this pattern
recognition methodology; the exercise provided a theoretical base for further
investigations and impetus for development work in this discipline. The obtained
8/3/2019 5 Keerthi Report New
27/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 27
results motivate the continuity of the system development considering a preprocessing
mechanism including normalization and slant removal.
Back propagation (BP) for training multilayer neural networks has many
shortcomings. Learning often takes insupportable time to converge, and it may fall
into local minima at all. One of the possible remedies to escape from local minima isusing a very small learning rate, but this will slow the learning process. Walid A.
Salameh and Mohammed A. Otair [32] proposed a system for the training of
multilayer neural networks with very small learning rate, especially when using large
training set size. It can apply in a generic manner for any network size that uses a
back propagation algorithm through optical time. They studied the performance of the
Optical Back propagation algorithm (OBP) on training a neural network for online
handwritten character recognition in comparison with back propagation.
3.1.5 Nearest-neighbor classifier
Among the various methods of supervised statistical pattern recognition, the Nearest
Neighbor rule achieves consistently high performance, without a priori assumptions
about the distributions from which the training examples are drawn. It involves a
training set of both positive and negative cases. A new sample is classified by
calculating the distance to the nearest training case; the sign of that point then
determines the classification of the sample. The k-NN classifier extends this idea by
taking the k nearest points and assigning the sign of the majority. It is common to
select k small and odd to break ties (typically 1, 3 or 5). Larger k values help reduce
the effects of noisy points within the training data set, and the choice of k is often
performed through cross-validation.
Abrita Chakravarty and William Day [33] proposed a recognition system for
handwritten digits. Nearest-neighbor (NN) and k-nearest neighbors (kNN) based
recognizers have widely been used for handwritten character recognition. When used
in applications, it is very important to compute reliable confidences corresponding to
the recognition results. The confidence values are typically computed during the post-
processing phase of the recognizer. They are the measures of correctness of output of
a recognizer. The estimation of confidences requires higher values to be assigned to
the correct recognition results, and lower values to the incorrect recognition results.
Vandana Roy and Sriganesh Madhvanath [34] have proposed A Skew-tolerant
Strategy and Confidence Measure for k-NN Classification of Online Handwritten
Characters. They explored the Adaptive-kNN strategy and confidence measure to
8/3/2019 5 Keerthi Report New
28/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 28
address the problem of online handwritten character recognition problem in presence
of skewed training sets. They showed through experiments, that performance of the
traditional kNN recognition strategy and confidence measure is highly sensitive to the
value of k, while adaptive-kNN strategy and confidence measure is not so sensitive to
the value of k. Hence, adaptive-kNN works well on the skewed training sets. Theyalso compared the performance of Adaptive-kNN strategy and related confidence
measure with various nearest neighbor based strategies and confidences, namely NN,
kNN, ACT, and observed that Adaptive-kNN strategy and confidence measure
outperforms the NN, kNN, and ACT, when the distribution of samples across classes
is skewed. This is a very promising technique for use in applications where the
distribution of training samples is skewed due to unbalanced data collection or due to
samples getting added over a period of use.
The k-nearest neighbor (kNN) rule is a simple and effective method for multi-way classification that is much used in Computer Vision. However, its performance
depends heavily on the distance metric being employed. Large margin nearest
neighbor (LMNN) classifier learns a distance metric for kNN classification and
thereby improves its accuracy. M. Pawan Kumar et. al., [35] extend the LMNN
framework to incorporate knowledge about invariance of the data. The main
contributions of their work are three fold: (i) Invariances to multivariate polynomial
transformations are incorporated without explicitly adding more training data during
learning - these can approximate common transformations such as rotations and
affinities; (ii) the incorporation of different regularizes on the parameters being learnt;
and (iii) for all these variations, they show that the distance metric can still be
obtained by solving a convex SDP problem. They call the resulting formulation
invariant LMNN (ILMNN) classifier.
3.2 Feature based schemes
Different features of handwritten scripts such as Structural, Syntactic, etc have been
used for recognition.
3.2.1 Transformation feature
In the online character recognition system for the handwritten Kannada characters,
proposed by Srinivasa Rao Kunte R and Sudhaker Samuel R D [36], wavelet features
are extracted from the contour of characters are used as features. The conventional
feed forward multilayer neural network is used as classifier. The results obtained are
8/3/2019 5 Keerthi Report New
29/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 29
most encouraging and the system can be extended to other similar Indian languages,
particularly, Telugu.
Rajput and Anita H. B [37] proposed a recognition system in which
recognition is based upon features extracted using Discrete Cosine Transform (DCT)
and Wavelets. The proposed method is experimented on handwritten documents ofeight Indian scripts that include English script and yielded encouraging results. The
above two listed system do not recognize numerals, to support this Diego et. al., [38]
proposed a recognition system which use Directional Continuous Wavelet Transform.
The percentage of correctly classified patterns was 99.17% and 90.20% for the
training set and the test set, respectively.
3.2.2 Structural feature
Online systems may also use features such as velocity, pressure, etc., that are captured
during writing. The temporal relations in online data are typically captured by
mathematical models like HMMs, Linear Prediction, etc., at the stroke or the sub
stroke level. Popular online handwriting recognition approaches give equal
importance to all parts of a stroke during matching, which may not be the best for all
cases. Karteek Alahari et. al., [39] proposed a recognition system which detect the
parts of a stroke (called sub stroke) that are more useful for the classification task.
Objective is to identify these sub strokes and use this information for improving the
performance of recognition schemes. Consider the problem of recognizing the
numerals 2 and 3 shown in Fig. 3.2.
Fig. 3.2: The sub strokes (sequence circled with dotted lines) of the numerals.
The two numerals appear to possess similar curvature properties at the
beginning of the sequences. As the complete numbers begin to appear, their
8/3/2019 5 Keerthi Report New
30/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 30
distinguishing characteristics unfold over time. In other words, the tail portion of the
numbers is more useful for distinguishing them. Authors described an approach to
identify critical segments of the strokes. The individual parts are then weighed to
obtain appropriate score for the final recognition.
The dataset consists of more than 1200 online numeral and character strokescollected from different people using an IBM CrossPad. To demonstrate the
applicability of their approach for discriminating two classes, they chose similar
character/ numeral pairs. To account for the variability in the data due to translation,
they normalize the features using a bounding box for the stroke and rescaling it to the
0-1 range. The accuracy of the classifiers reported here are much lower than the
commercial or most of the reported recognition systems. This is due to the fact that (a)
datasets were not tuned to achieve higher recognition or preprocessed to suit a
specific recognition scheme, and (b) implementation of HMMs, DTW, etc. is nottuned for the online data case.
Aparna et. al., [40] proposed a system for online recognition of handwritten
Tamil characters in which handwritten character is constructed by executing a
sequence of strokes. A structure or shape-based representation of a stroke is used in
which a stroke is represented as a string of shape features. Using this string
representation, an unknown stroke is identified by comparing it with a database of
strokes using a flexible string matching procedure.
3.2.3 Statistical feature
The recognition system reported by R. J. Ramteke [41] makes use of Invariant
Moments feature for handwritten Devanagari vowels recognition. The system is
independent of size, slant, orientation, translation and other variations in handwritten
vowels. In order to enhance the performance of the system, an attempt has been made
to compute invariant moments by small perturbation in image and information is
extracted from the perturbation. But it was found that, another local feature descriptor,
image partition in different zoning is better representation of the features than
perturbation. The Fuzzy Gaussian Membership function has been adopted for
classification. The success rate of the method is found to be 94.56%.
Hiroto Mitoma et. al., [42]proposed online character recognition system based
on Elastic Matching and Quadratic Discrimination to overcome the over fitting
problem which often degrades the performance of elastic matching based online
character recognizers. In the proposed technique, elastic matching is used just as an
8/3/2019 5 Keerthi Report New
31/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 31
extractor of a feature vector representing the difference between input and reference
patterns. They have obtained 97.95% accuracy.
3.2.4 Local features
Prasanth et. al., [43] proposed a character based elastic matching using local features
for recognizing online handwritten data. Dynamic Time Warping (DTW) has beenused with four different feature sets: x-y features, Shape Context (SC) and Tangent
Angle (TA) features, Generalized Shape Context feature (GSC) and the fourth set
containing x-y, normalized first and second derivatives and curvature features.
Nearest neighborhood classifier with DTW distance was used as the classifier. In
comparison, the SC and TA feature set was found to be the slowest and the fourth set
was best among all in the recognition rate. The results have been compiled for the
online handwritten Tamil and Telugu data. On Telugu data they obtained an accuracy
of 90.6% with a speed of 0.166 symbols/sec. To increase the speed they haveproposed a 2-stage recognition scheme using which they obtained accuracy of 89.77%
but with a speed of 3.977 symbols/sec.
3.2.5 Star feature
The features extracted from the character should encode the local, global and the
structural characteristics of the character shape. Dinesh M & Murali Krishna Sridhar
[44] proposed A Feature based on Encoding the Relative Position of a Point in the
Character for Online Handwritten Character Recognition. They proposed a new
feature for recognition of online handwritten characters called the star feature. The
star feature encodes the local, global and structural characteristics of a character. The
star feature describes every point of the character, in terms of its relative position with
respect to the other points in the character. The experimental results show that the star
feature achieves high accuracy on both the data sets.
8/3/2019 5 Keerthi Report New
32/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 32
CHAPTER-4
SUMMARY
It is found from the literature that recognition of characters from Indian languages in
general is more difficult than for European languages because of the large number of
vowels, consonants, and conjuncts (combination of vowels and consonants).
Following are the findings obtained from the literature survey on different
classification schemes for on-line handwriting recognition.
DTW classifier requires an adaptation framework which is particularly useful
for rapidly deploying recognizers for new scripts. The memory requirements for the
Active-DTW recognizer and time taken for adaptation were shown to be significantly
less as compared to the nearest neighbor classifier. The combination of subspace and
DTW classifiers, even though compatible all the three modes, namely, writer
dependent, writer independent and writer adaptive, is not suitable for commercial use
because that is only to reap the advantages of both the methods. Prototype based
DTW classifier is more suitable for automatic recognition systems.
SVM classifier used for online handwriting recognition of Indian script has
been found to give the lowest recognition accuracy for set of baseline strokes that
loops. SVM with features from Functional Approximation results in good accuracy
for mathematical symbols. A data-driven HMM-based online handwritten word
recognition system obtained accuracies ranged from 98% to 92.2% with different
lexicon sizes. The relatively low performance in the case of high lexicon size can be
improved by the use of statistical language models, which are commonly applied in
Western cursive recognition.
Counter propagation Neural Net used as classifiers is found to be more
economical than other NN architectures such as back-propagation where the training
time can take long time but multilayer neural networks with very small learning rate
can be developed using Optical Back propagation.
In case of Nearest-neighbor classifier, performance of the traditional kNN
recognition strategy and confidence measure is highly sensitive to the value of k,
while adaptive-kNN strategy and confidence measure is not so sensitive to the value
of k. Hence, adaptive-kNN works well on the skewed training sets. However, its
performance depends heavily on the distance metric being employed, so Large margin
8/3/2019 5 Keerthi Report New
33/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 33
nearest neighbor (LMNN) classifier can improve the accuracy of recognition system
by learning a distance metric for kNN classification.
Wavelet feature based recognition system developed for Kannada characters
can be extended to other similar Indian languages, particularly, Telugu. Discriminant
sub strokes based system improves the performance of feature based recognitionschemes. Star feature which is based on encoding the relative position of a point in the
character has been found to encode the local, global and the structural characteristics
of the character shape there by providing high accuracy.
8/3/2019 5 Keerthi Report New
34/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 34
REFERENCES
[1] Anuj Sharma, R. Kumar and R.K. Sharma, HMM based Online HandwrittenGurmukhi Character Recognition, International Journal of Machine Graphics
and Vision, 2010.
[2] Santosh K.C, Cholwich Nattee., A Comprehensive Survey on On-lineHandwriting Recognition Technology and its real application to the Nepalese
handwriting, Kathmandu University Journal of Science, Vol. 5, No. 1, Jan,
2009, pp 31-55.
[3] Tappert, C. C., Suen, C. Y., Wakahara, T., 1990., The state of the art in onlinehandwriting recognition, IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 12, no. 8, pp. 787-808.
[4] Tappert, C. C, Adaptive online handwriting recognition. In Proceedings ofInternational Conference of Pattern Recognition, pp. 1004-1007, 1984.
[5] Kuklinski, T. T., Components of handprint style variability. In Proceedings ofSeventh International Conference of Pattern Recognition, pp. 924-926, 1984.
[6] Beigi, H., Nathan, K., Clary, G. J., and Subhramonia, J., Size normalization inunconstrained online handwritng recognition, In Proceedings ICIP, pp. 169-
173, 1994.
[7] Subrahmonia, J, Zimmerman, T., Pen computing: challenges and applications.In Proceedings of 15th International Conference on Pattern Recognition, vol. 2,
pp. 60-66, 2000.
[8] Unser, M., Aldroubi, A., Eden, M., B-Spline signal processing: part II - efficientdesign and applications. IEEE Transactions on Signal Processing, vol. 41, no. 2,
pp. 834-848, 1993.
[9] Bozinovic, R. M., Srihari, S. N., Offline cursive script recognition, IEEETransactions on Pattern Analysis and Machine Intelligence, vol. 2, no. 1, pp. 68-
83, 1989.
[10] Guillevic, D. and Suen, C. Y., Cursive script recognition- a sentence levelrecognition scheme, In Proceedings of IWFHR, pp. 216-223, 1994.
[11] Negi, A., Swaroop, K. S., Agarwal, A., A correspondence based approach tosegmentation of cursive words, In Proceedings of International Conference on
8/3/2019 5 Keerthi Report New
35/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 35
Document Analysis and Recognition, pp. 1034-1037, 1995.
[12] Simoncini, L., Kovacs, M., A system for reading USA Census 90 handwrittenfields, In Proceedings of International Conference on Document Analysis and
Recognition, vol. 2, pp. 86-91, 1995.
[13] Brault, J. J. and Plamondon, R., Segmenting handwritten signatures at theirperceptually important points, IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 15, no. 9, pp. 953-957, 1993.
[14] Oivind due Trier, Anil K. Jain, Torfinn Taxt, Feature extraction methods forcharacter recognition a survey, Michigan state University, East Lansing,
1995.
[15] Kavallieratou, E., Fakatakis, N., Kolkkinakis, G., An unconstrained handwritingrecognition system. International Journal of Document Analysis and
Recognition, vol. 4, no. 4, pp. 226-242, 2002.
[16] Anil K. Jain, Statistical pattern recognition: A review, IEEE Transactions onPattern Analysis and Machine Intelligence, vol. 22, no. 1, 2000.
[17] Plamondon, R. and Srihari, S. N., Online and offline handwriting recognition: Acomprehensive survey. IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 22, no. 1, pp. 63-84, 2000.
[18] Uchida, S. and Sakoe, H., A survey of elastic matching techniques forhandwritten character recognition. IEICE Trans. Information and Systems,
volE88-D, no. 8, pp. 1781-1790, 2005.
[19] M. Sridhar, D. Mandalapu, and M. Patel. Active- DTW: A generative classierthat combines elastic matching with active shape modeling for online
handwritten character recognition. International Conference on Frontiers in
Handwriting Recognition, 99(7):1.100, November 1999.
[20] Vandana Roy, Sriganesh Madhvanath, Anand S., Raghunath R. Sharma, AFramework for Adaptation of the Active-DTW Classifier for Online Handwritten
Character Recognition, International Conference on Document Analysis and
Recognition, Barcelona, August 2009.
[21] Niranjan Joshi, G.Sita, A.G.Ramakrishnan and S.Madhvanath, Comparison ofelastic matching algorithms for online Tamil handwritten character recognition,
8/3/2019 5 Keerthi Report New
36/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 36
In Proc. IWFHR-9, Tokyo, Japan, Oct. 26-29, 2004, pp. 444-449.
[22] Ralph Niels and Louis Vuurpij, Dynamic Time Warping Applied to TamilCharacter Recognition, In proceedings of Eighth International Conference on
Document Analysis and Recognition, August 31 to September 1, 2005, Seoul,
Korea
[23] Muzaffar Bashir, and Jurgen Kempf, Bio-inspired reference level assignedDTW for person identification using handwritten signatures, Proceedings of the
2009 joint COST 2101 and 2102 international conference on Biometric ID
management and multimodal communication, Madrid, Spain, 2009.
[24] Abdul Rahim Ahmad, Christian Viard-Gaudin, Marzuki Khalid, and RubiyahYusof, Online Handwriting Recognition using Support Vector Machine,
Proceedings of the Second International Conference on Artificial Intelligence in
Engineering & Technology, August 3-5 2004, Kota Kinabalu, Sabah, Malaysia.
[25] Swethalakshmi H, Jayaraman A, Chakravarthy V.S., and Sekhar C.C., OnlineHandwritten Character Recognition of Devnagari and Telugu Characters using
Support Vector Machines, Proceedings of 10th
IWFHR, 2006.
[26] Birendra Keshari and Stephen M. Watt, Online Mathematical SymbolRecognition using SVMs with Features from Functional Approximation, Ninth
International Conference on Document Analysis and Recognition, vol. 2, pp 859-
863, 2007.
[27] Bharath A, Sriganesh Madhvanath Hidden Markov Models for OnlineHandwritten Tamil Word Recognition HP Laboratories India, HPL-2007-108,
July 6, 2007.
[28] Jagadeesh Babu V, Prasanth L, Raghunath Sharma R, Prabhakara Rao G.V. andBharath A, HMM-based Online Handwriting Recognition System for Telugu
Symbols,ICDAR,2007, 23-26 September 2007, Curitiba, Brazil.
[29] U. Bhattacharya, B.K.Gupta, S.K. Parui, Direction Code Based Features forRecognition of Online Handwritten Characters of Bangla, Proc. 9th ICDAR,
vol.1, pp. 58 - 62, 2007.
[30] Ahmed, S.M.,et.al,. Experiments in character recognition to develop tools for anoptical character recognition system, IEEE Inc. 1st National Multi Topic Conf.
proc. NUST, Rawalpindi, Pakistan, 61-67, Nov.1995.
8/3/2019 5 Keerthi Report New
37/38
Online Handwriting Recognition
Dept. Of CSE, JNNCE Page 37
[31] Muhammad Faisal Zafar, Dzulkifli Mohamad, and Razib M. Othman. On-lineHandwritten Character Recognition: An Implementation of Counter propagation
Neural Net, 2004.
[32] Walid A. Salameh and Mohammed A. Otair, Online Handwritten CharacterRecognition Using an Optical Backpropagation Neural Network, Southampton
Institute, pp. 33-52, 1999.
[33] Abrita Chakravarty and William Day, Handwritten Digit Classification,Michigan State University, East Lansing, 1995.
[34] Vandana Roy, Sriganesh Madhvanath, Anand S., Raghunath R. Sharma, AFramework for Adaptation of the Active-DTW Classifier for Online Handwritten
Character Recognition, International Conference on Document Analysis and
Recognition, Barcelona, August 2009.
[35] M. Pawan Kumar, P.H.S. Torr and A. Zisserman, An Invariant Large MarginNearest Neighbour Classifier, IEEE 11th International Conference on ICCV,
2007.
[36] Srinivasa Rao Kunte, R, Sudhaker Samuel, R D On-line character recognitionfor handwritten Kannada characters using wavelet features and neural classifier
IETE J RES. Vol. 46, no. 5, pp. 387-392. 2000.
[37] Rajput and Anita H. B, Recent Trends in Image Processing and PatternRecognition, RTIPPR, 2010.
[38] Diego J. Romero, Leticia M, Seijas and Ana M. Ruedin, Directional ContinuousWavelet Transform Applied to Handwritten Numerals Recognition Using Neural
Networks, JCS&T Vol. 7 No. 1, 2007.
[39] Karteek Alahari, Satya Lahari Putrevu, C. V. Jawahar, "Discriminant Substrokesfor Online Handwriting Recognition," Eighth International Conference on
Document Analysis and Recognition (ICDAR'05), pp.499-505, , 2005.
[40] K.H. Aparna, Vidhya Subramanian, M. Kasirajan, G. Vijay Prakash, V.S.Chakravarthy, Online Handwriting Recognition for Tamil, In Proceedings of
the Ninth International Workshop on Frontiers in Handwriting Recognition,
Pages: 438 443, 2004
[41] Ramteke, Invariant Moments Based Feature Extraction for Handwritten
8/3/2019 5 Keerthi Report New
38/38
Online Handwriting Recognition
Devanagari Vowels Recognition, International Journal of Computer
Applications 1(18):15, February 2006.
[42] Hiroto Mitoma, Seiichi Uchida, and Hiroaki Sakoe, Online CharacterRecognition Based on Elastic Matching and Quadratic Discrimination, Kyushu
University, Japan, 2005.
[43] L.Prasanth,V.Jagadeesh Babu, R. Raghunath Sharma, G.V.Prabhakara Rao, andDinesh M, Elastic Matching of Online Handwritten Tamil and Telugu Scripts
Using Local Features, In Proceedings of ICDAR, 2007.
[44] Dinesh M, Murali Krishna Sridhar, A Feature based on Encoding the RelativePosition of a Point in the Character for Online Handwritten Character
Recognition,ICDAR Ninth International Conference, 2007.