Advance in Electronic and Electric Engineering.
ISSN 2231-1297, Volume 4, Number 6 (2014), pp. 555-560
© Research India Publications
http://www.ripublication.com/aeee.htm
Performance Analysis of Various Classifier Technique
Used in Handwritten Digit Recognition
Kunal J. Patil and Anil H. Karode
SSBT’s Coet, Bambhori, Jalgaon (MS)
Abstract
Accelerometer based gesture recognition is one of the widely
implemented method in the recognition scenario. We have
implemented a 3D input digital pen which works on triaxial
accelerometer to sense human gesture. This digital pen embedded with
triaxial accelerometer, microcontroller, RF wireless transmitter
module. The triaxial accelerometer measure acceleration signal along
all the 3 axis. Accelerated signal process through microcontroller and
serially transmitted through RF transmitter which can be received at
remote place RF receiver. With the help of MATLAB tool feature
vector are generated from received accelerated signal to recognize
handwritten numeric digit and various classifier technique for the best
accuracy purpose. Our experimental results shows that the PNN has
best accuracy than any other classifier.
Keyword: MEMS,PNN,SVM,HMM.
1. Introduction
Handwriting Recognition is generally used for security ,authentication purpose. There
are two types recognition offline recognition & on-line recognition. The projected
system is a web hand writing character recognition. The character recognition is
finished by AN MEMS measuring instrument. This accelerometer may be a 3D axial
digital output and offers response for each slight deflection or movement within the
system. it's developed by victimization MEMS technology.(2)A vital advantage of
measuring instrument for general motion sensing is that they will be operated with
none external reference and limitation in operating conditions. However, gesture
recognition is comparatively sophisticated as a result of completely different
completely users have different speeds and designs to get varied motion trajectories.
556
Kunal J. Patil & Anil H. Karode
Thus, several researchers have tried to increasing the accuracy of handwriting
recognition systems.(1)
Recently, some researchers have focused on reducing the error of handwriting
mechanical phenomenon reconstruction by manipulating acceleration signals and
angular velocities of mechanical phenomenon sensors. However, the reconstructed
trajectories suffer from varied intrinsic errors of mechanical phenomenon sensors.
Hence, several researchers have centered on developing effective algorithms to cut
back error of mechanical phenomenon sensors & to enhance the popularity
accuracy. (2)
AN economical acceleration error compensation formula supported zero speed
compensation was developed to cut back acceleration errors for deed correct
reconstructed mechanical phenomenon. The options of the preprocessed acceleration
signals of every axis embrace mean, correlation among axes, inter quartile vary (IQR),
mean absolute deviation (MAD), root mean sq. (rms), VAR, variance (STD), and
energy. Before classifying the hand motion trajectories, we have a tendency to perform
the procedures of feature choice and extraction strategies AN extended Kalman filter
with magnetometers (micro mechanical phenomenon measure unit (μIMU with
magnetometers) was utilized to compensate the orientation of the projected digital
writing instrument. If the orientation of the instrument was calculable exactly, the
motion trajectories of the instrument were reconstructed accurately(1) .
2. Hardware Design of Digital Pen
Fig. 1: Hardware Design of Digital Pen.
Digital pen carries with it microcontroller, RF transreceiver & Triaxial
measuring system. The triaxial measuring system measures the acceleration signals
generated by a user’s hand motions. The microcontroller collects the analog
acceleration signals and converts the signals to digital ones via the A/D convertor. The
wireless transceiver transmits the acceleration signals wirelessly to a PC. The output
signals of the measuring system are sampled at one hundred cycles/second by the 12-b
A/D convertor. Then, all the information detected by the measuring system are
Performance Analysis of Various Classifier Technique Used in Handwritten
557
transmitted wirelessly to a computer by AN RF transceiver at a pair of.4-GHz
transmission band with 1-Mb/s transmission rate[B].The schematic of hardware style
of digital pen as shown in fig1.
3. Trajectory Recognition Algorithm
A mechanical phenomenon is that the lane that a moving object follows through area
as a job of your time. A mechanical phenomenon may be a sequence ( f k ( x)) x  N of
values thought of by the iterated application of a mapping f to a part x of its supply.
.
Fig. 2: Illustration showing the trajectory of a packet at an uphill transceiver target.
The diagram of the projected mechanical phenomenon recognition algorithmic
program consisting of acceleration acquisition, signal preprocessing, feature
generation, feature choice, and have extraction is shown fig three.
Fig. 3: Block Diagram of Trajectory recognition algorithm[1]
558
Kunal J. Patil & Anil H. Karode
3.1 Flow of Trajectory Recognition algorithm
The raw acceleration signals of hand motions are generated by the measuring device
and picked up by the microcontroller. as a result of attribute, our hand continually
trembles slightly whereas moving, that causes certain quantity of noise. The signal
preprocessing consists of standardization, a moving average filter, a high-pass filter,
and normalization (1). as a result of moving average filter the signal become electric
sander and explosive amendment in signal as a result of hand movement is avoided
with the assistance of high pass filter. The normalization is begin to start out to begin
the signal from start purpose. The characteristics of various hand movement signals are
often obtained by extracting options from the preprocessed x,y, and coordinate axis
signals, and that we extract eight options from the triaxial acceleration signals, together
with mean, STD, VAR, IQR, correlation between axes ,MAD, rms, and energy .(1)
Feature choice includes a variety criterion and a pursuit strategy. The principally used
choice criterion is that the KBCS that is originally developed by Wang .For feature
extraction there ar numerous sort of numerous sort of classifier like PNN , FNN, FDA,
HMM, GMM,LDA. they need totally different recognition rate.
4. Different Classifier for Recognition
4.1 PNN Classifier
The PNN was initial projected by Specht , the PNN is certain to converge to a
Bayesian classifier, and thus, it's a good potential for creating classification selections
accurately and providing chance and responsibility measures for every classification.
additionally, the procedure of the PNN solely wants one epoch to regulate the weights
and biases of the spec. Therefore, the foremost necessary advantage of mistreatment
the PNN is its high speed of learning. Typically, the PNN consists of associate input
layer, a pattern layer, a summation layer, and a choice layer as shown in Fig. 4. The
operate of the neurons in every layer of the PNN is outlined as follows. is defined as
follows.
1) Layer 1: the primary layer is that the input layer, and this layer performs no
computation. The neurons of this layer convey the input options x to the
neurons of the second layer
2) Layer 2: The second layer is that the pattern layer, and therefore the variety of
neurons during this layer is adequate to NL.
3) Layer 3: The third layer is that the summation layer. The contributions for
every category of inputs are summed during this layer to provide the output
because the vector of possibilities. (2) where Ni is that the total variety of
samples within the kth nerve cell.
4) Layer 4: The fourth layer is that the call layer
Performance Analysis of Various Classifier Technique Used in Handwritten
559
Fig. 4: Topology of PNN classifier [1]
4.2 HMM Classifier
Markov model may be a mathematical model of framework wherever these processes
generate a random sequence of outcomes per sure possibilities (3,4). it's trainable and
therefore the underlying framework is unperceivable, therefore we have a tendency to
decision it hidden Markov model. each CHMM and DHMM square measure wide
employed in Pattern Recognition and lots of different fields. For DHMM, the evident
states and invisible states square measure all separate.
A HMM may be a assortment of finite states S = interconnected by transitions.
every state includes a variety of distinct observation symbols V = reminiscent of the
physical output of the system (2)
4.3 FDA Classifier
FDA is one of the linear projection methods that project the input point (a vector) in
the input space to a point in the feature space. One motivation of using a linear method
was that the training is easier, faster and requires relatively smaller amount of data for
reasonable level of training than the more resource-intensive techniques like neural
networks or hidden Markov models. Therefore it expedites, as a fast-running test-bed,
one of our purposes, which is to explore the various sensor information combinations
and see how the classifier behaves on each combination. One reason for such an
exploration was that we wanted to determine the best performing combination of
sensors. Another reason was to identify the most economical alternatives (yet
performing acceptably) in terms of the number of sensors because the less sensors we
use, the cheaper. Another motivation for linear method was to reinforce the overall
560
Kunal J. Patil & Anil H. Karode
performance via an ensemble of simple and fast classifiers. Yet another motivation was
that the approach has a potential for making a user-tailored adaptation feasible because
the training runs fast and demands less on the amount of training data.
Table 1: Recognition rate of Different Classifier[1,3,5]
Classifier
HMM
PNN
FDA
FNN
Recognition Rate
96.2 %
98.75%
93.23%
96.25%
5. Conclusion
This paper gives construction of a digital pen that is beneficial in security purpose
supported mechanical phenomenon recognition formula. The accuracy of pen depends
upon the the utilization of classifier from study we tend to come back to conclude that
the popularity rate of probabilistic neural network is bigger than different classifier and
this pen are often used wherever robust security is needed.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
Jeen-Shing Wang and Fang-Chen Chuang, “An Accelerometer-Based Digital
Pen With a Trajectory Recognition Algorithm for Handwritten Digit and
Gesture Recognition,” IEEE Sens. J., vol. 59, no. 07, pp. 1543–1551, July.
2012.
Z. Dong, U. C. Wejinya, and W. J. Li, “An optical-tracking calibration method
for MEMS-based digital writing instrument,” IEEE Sens. J., vol. 10, no. 10,
pp. 1543–1551, Oct. 2010.
J. S.Wang, Y. L. Hsu, and J. N. Liu, “An inertial-measurement-unit-based pen
with a trajectory reconstruction algorithm and its applications,” IEEE Trans.
Ind. Electron., vol. 57, no. 10, pp. 3508–3521, Oct. 2010.
S.-H. P. Won, W. W. Melek, and F. Golnaraghi, “A Kalman/particle filterbased position and orientation estimation method using a position
sensor/inertial measurement unit hybrid system,” IEEE Trans. Ind. Electron.,
vol. 57, no. 5, pp. 1787–1798, May 2010.
A. D. Cheok, Y. Qiu, K. Xu, and K. G. Kumar, “Combined wireless hardware
and real-time computer vision interface for tangible mixed reality,” IEEE
Trans. Ind. Electron., vol. 54, no. 4, pp. 2174–2189, Aug. 2007.
E. Sato, T. Yamaguchi, and F. Harashima, “Natural interface using pointing
behavior for human–robot gestural interaction,” IEEE Trans. Ind. Electron.,
vol. 54, no. 2, pp. 1105–1112, Apr. 2007.
Y. S. Kim, B. S. Soh, and S.-G. Lee, “A new wearable input device:
SCURRY,” IEEE Trans. Ind. Electron., vol. 52, no. 6, pp. 1490–1499, Dec.
2005.