Surface EMG pattern recognition for real-time control of a wrist exoskeleton
© Khokhar et al; licensee BioMed Central Ltd. 2010
Received: 4 May 2010
Accepted: 26 August 2010
Published: 26 August 2010
Surface electromyography (sEMG) signals have been used in numerous studies for the classification of hand gestures and movements and successfully implemented in the position control of different prosthetic hands for amputees. sEMG could also potentially be used for controlling wearable devices which could assist persons with reduced muscle mass, such as those suffering from sarcopenia. While using sEMG for position control, estimation of the intended torque of the user could also provide sufficient information for an effective force control of the hand prosthesis or assistive device. This paper presents the use of pattern recognition to estimate the torque applied by a human wrist and its real-time implementation to control a novel two degree of freedom wrist exoskeleton prototype (WEP), which was specifically developed for this work.
Both sEMG data from four muscles of the forearm and wrist torque were collected from eight volunteers by using a custom-made testing rig. The features that were extracted from the sEMG signals included root mean square (rms) EMG amplitude, autoregressive (AR) model coefficients and waveform length. Support Vector Machines (SVM) was employed to extract classes of different force intensity from the sEMG signals. After assessing the off-line performance of the used classification technique, the WEP was used to validate in real-time the proposed classification scheme.
The data gathered from the volunteers were divided into two sets, one with nineteen classes and the second with thirteen classes. Each set of data was further divided into training and testing data. It was observed that the average testing accuracy in the case of nineteen classes was about 88% whereas the average accuracy in the case of thirteen classes reached about 96%. Classification and control algorithm implemented in the WEP was executed in less than 125 ms.
The results of this study showed that classification of EMG signals by separating different levels of torque is possible for wrist motion and the use of only four EMG channels is suitable. The study also showed that SVM classification technique is suitable for real-time classification of sEMG signals and can be effectively implemented for controlling an exoskeleton device for assisting the wrist.
sEMG can provide information regarding the neural activation of muscles, which can be used to estimate the intention of the person and also identify potential neuromuscular disorders . The use of sEMG signals has been explored for different applications. One of the applications of sEMG signals is in regards to rehabilitation through robotic devices. It has been proposed that sEMG signals can be used to quantify the assessment of hand functions  and robotic devices can be used to provide an assistive force as a compensation for hand movement . Combining sEMG signals with robotic therapy can optimize the coordination of motor commands and actual movement [4–6]. Another application of EMG signals is in the control of prosthetic hands. Numerous prosthetic hands have been prototyped, including the CyberHand  and SmartHand , and some have also been commercialized, including the iLimb  and the Otto Bock's SensorHand Speed . In these research and development efforts, the goal was to obtain a lightweight and dexterous prosthetic hand that could perform movements similar to a human hand. A crucial aspect towards an effective use of these prosthetic hands is their intuitive control, which could be achieved through detection and interpretation of the user's neurological activity to be detected, for example, through sEMG electrodes. Whether used for controlling an assistive, rehabilitative or prosthetic device, the basic challenge is to be able to process sEMG signals and identify the intention of the user. Different studies have been performed to tackle this challenge by using different pattern recognition methods [11–28].
The analysis of pattern recognition in sEMG mainly consists of two steps, namely feature extraction and classification. Feature extraction is the dimensionality reduction of the raw sEMG input to form a feature vector - the accuracy of the pattern classification system almost entirely depends on the choice of these features . Features cannot be extracted from the individual samples as the structural detail of the signal will be lost and hence the features need to be calculated by segmenting the raw sEMG signal and calculating a set of features from each segment . Researchers have experimented with the length of the segment and the constraint in the length mainly derives from the specific real-time implementation. A delay of 200~300 ms interval is the clinically recognized maximum delay tolerated by the users . A suitable delay for the controller to generate a control command should therefore be between 100~125 ms . Different features have been used in pattern recognition involving both time domain and time-frequency domain features. Some of these include mean absolute value [11, 12, 15–17], zero crossings (ZC) [11, 12, 15–17], slope sign changes (SSC) [11, 12, 15, 16], autoregressive (AR) model coefficients [12, 15, 18–20], cepstrum coefficients , waveform length (WL) [11, 12, 16, 17] and wavelet packet transform[13–15].
As regards to classification, it can be defined as the process of assigning one of K discrete classes to an input vector x. Numerous studies have been done to classify the features extracted from the sEMG like neural networks [11, 20, 21], bayesian classifier , linear discriminant analysis [16, 23], hidden markov model , multilayer perceptron [13, 14, 23], fuzzy classifier [15, 17–19], gaussian mixture model  and support vector machines (SVM) [21, 22, 27, 28].
Feature extraction and classification methods were primarily used in previous research studies to identify the orientation of the hand without considering the amount of force the user was applying. In the use of advanced hand prostheses, it would however be beneficial having control over the amount of force a person intends to apply and, for assistive devices, force control would indeed be necessary. Castellini et al.  successfully controlled the amount of force applied by the fingers in different types of grasp so that the user could apply a different amount of force for holding, for example, a hammer or an egg .
Support Vector Machines
where N is the number of data points, x n is the vector representing a data point, t n is the label associated with a data point, y is the learned model, w is the vector representing adaptive model parameters, ξ n is the slack variable and C > 0 is the penalty factor. Although SVM linearly separates two data sets, different researchers have introduced the use of kernels in the algorithm to extend it for non-linear separation without much increase in computational complexity. Some of the well-known kernels include polynomial, radial basis, Gaussian and sigmoid. SVM, which is a two class separation technique, has also been extended for multiclass classification. This is done by splitting a single multi-class problem to multiple binary classification problems. The two most common methods are one-versus-one and one-versus-all, whose details are presented in . An important property of SVM is that the model parameter estimation corresponds to a convex optimization problem meaning that any local solution will be a global optimum . SVM also has a high generalization ability making it suitable for unseen data; it has recently been successfully applied to bio-information signals for pattern recognition [34–37].
EMG electrode placement and data acquisition
Reliable sEMG data acquisition is necessary before extracting features for classification. Numerous factors affect the quality of sEMG acquisition such as inherent noise in the electronic equipment, ambient noise in the surrounding atmosphere, motion artefacts and poor contact with skin. The first three factors are dependent on the sEMG acquisition system used and, to reduce the effects of these, a commercial sEMG system from Noraxon (Myosystem 1400L) was used. In order to have a good skin contact with the electrodes, the guidelines of the surface electromyography for the non-invasive assessment of muscles (SENIAM) project  were followed. The skin of the volunteer was shaved and an alcohol swab was used to clean the skin. The electrodes were placed at the desired locations after the skin dried. We used AgCl gel dual electrodes from Noraxon, which contains two electrodes at a recommended distance. The usable energy in an EMG signal lies in the range of 0-500 Hz  and therefore the acquired sEMG signal was digitized at 1024 samples per second using a data acquisition card from National Instruments (NI USB-6289) and stored on a computer by the LabVIEW software.
Data collection setup and protocol
Number of Repetition
Wrist flexion with maximum torque
Wrist extension with maximum torque
Wrist flexion: start from rest and increase torque by 10% of MVC after every 10 seconds until 50% of MVC is applied
Wrist flexion: start from 50% of MVC and decrease torque by 10% after every 10 seconds until no torque is applied
Wrist extension: start from rest and increase torque by 10% of MVC after every 10 seconds until 50% of MVC is applied
Wrist extension: start from 50% of MVC and decrease torque by 10% after every 10 seconds until no torque is applied
Wrist ulnar deviation with maximum torque
Wrist radial deviation with maximum torque
Wrist ulnar deviation: start from rest and increase torque by 10% of MVC after every 10 seconds until 40% of MVC is applied
Wrist ulnar deviation: start from 40% of MVC and decrease torque by 10% after every 10 seconds until no torque is applied
Wrist radial deviation: start from rest and increase torque by 10% of MVC after every 10 seconds until 40% of MVC is applied
Wrist radial deviation: start from 40% of MVC and decrease torque by 10% after every 10 seconds until no torque is applied
Feature extraction and classification
where x i is the value of the i th sample in the k th segment and N is the number of samples, which in our case is 256.
where a i are the model coefficients, m is the order of the model and ε is the output error. We used the AR model coefficients as the features with a model order of four, which is adequate for modelling EMG signals , thus generating four features for each channel of sEMG.
We used eight fold cross validation along with grid search to find the optimal parameters for C and γ.
Actions for different classes
Flexion with 10% of MVC torque
Flexion with 20% of MVC torque
Flexion with 30% of MVC torque
Flexion with 40% of MVC torque
Flexion with 50% of MVC torque
Extension with 10% of MVC torque
Extension with 20% of MVC torque
Extension with 30% of MVC torque
Extension with 40% of MVC torque
Extension with 50% of MVC torque
Ulnar deviation with 10% of MVC torque
Ulnar deviation with 20% of MVC torque
Ulnar deviation with 30% of MVC torque
Ulnar deviation with 40% of MVC torque
Radial deviation with 10% of MVC torque
Radial deviation with 20% of MVC torque
Radial deviation with 30% of MVC torque
Radial deviation with 40% of MVC torque
Mechanical design and control of exoskeleton
The flexion/extension motion is provided by a linear actuator, having 10 cm stroke length (Firgelli L12-100-210-12-P), which is fixed to a moveable housing coupled to an arc-shaped disk of the forearm brace, as shown in Figure 4. The head of the linear actuator is connected to a block having two aluminium square rod extensions used to improve the stiffness of the WEP during actuation. Two parallel bars are attached to connect the aluminium extensions with the hand brace through revolute joints. The linear actuator is able to deliver about 2.2 Nm of torque to the wrist over the entire flexion-extension range of motion when supplied with 12 V.
To control the ulnar/radial deviation of the wrist, a gear motor (Pololu 298:1 micro metal gear motor) is attached to a side of the linear actuator housing, and coupled to the outer side of the arc-shape disk with a spur gear. The ratio between the arc-shape disk's radius and the one of the spur gear is 15:1; thus, the torque generated by the gear motor is amplified by a factor of 15 at the wrist joint. With the use of the Pololu gear motor, a maximum torque of 5.4 Nm can be applied at the wrist joint for ulnar/radial deviation.
Real-time experimental setup
The real-time experiment consisted of two steps: training and testing. During the first step, the volunteer was asked to place the right forearm on the custom made rig, which indicated the torque applied by the user in real-time. The sEMG acquisition system, presented in the data acquisition section of this paper, was used. The torque and EMG data were digitalized at a frequency of 1024 samples per second. The volunteer applied the torque according to the proposed protocol (see Table 1) and 13 classes were trained. In the second step, the volunteer applied different torques by using the same setup and the LabVIEW application predicted the wrist output through the only real time sEMG input and provided the control signal to actuate the WEP, which applied torque corresponding to the identified class.
Wrist assistance: proof of concept
The purpose of the test was to enable a comparison between the rms values of the sEMG with and without the WEP assistance. The overall experiment consisted of three steps: (1) training for the classification system, (2) wrist extension with assistance from the WEP and (3) wrist extension without assistance. During the training step, the parallel bars of the WEP were detached from the hand brace so that the wrist was not constrained and the force sensor could read the applied force. The classification system was then trained for four classes corresponding to rest, 10% of MVC, 20% of MVC and 30% of MVC. In the next step, the parallel bars of the WEP were attached back to the WEP to assist the wrist extension. The volunteer was asked to pull against the force sensor, and maintain a strength that corresponded to a particular class for a short period - the WEP was expected to assist the wrist extension. In the last step, the parallel bars of the WEP were detached again from the hand brace to remove the assistance. The volunteer was subsequently asked to pull against the force sensor to a force level that was achieved with assistance, and maintain that force level for a short period of time - visual feedback of the applied force was provided to the volunteer.
Results and Discussion
Classification results with 19 classes
Cross Validation Accuracy (%)
Testing Accuracy (%)
Volunteer # 1
Volunteer # 2
Volunteer # 3
Volunteer # 4
Volunteer # 5
Volunteer # 6
Volunteer # 7
Volunteer # 8
Results obtained for classification accuracy in volunteers who had greater MVC and those who could maintain a torque level with little variation were much better than the rest. Also, most of the errors were due to a class misclassified in an adjacent class. The average accuracy for the eight volunteers neglecting misclassification in adjacent classes reached up to 99.99%. This suggests that the cause of lower accuracy is the small separation between torque levels; to evaluate the trade-off between smoothness of torque and average accuracy of the classifier, the second configuration was analyzed.
Classification results with 13 classes
Cross Validation Accuracy (%)
Testing Accuracy (%)
Volunteer # 1
Volunteer # 2
Volunteer # 3
Volunteer # 4
Volunteer # 5
Volunteer # 6
Volunteer # 7
Volunteer # 8
Tables 3 and 4 show that, as expected, classification accuracy decreased when the number of classes increased but still good results were obtained with the highest number of classes. Depending on the needs of specific future practical applications, which could have different requirements on the smoothness of the output torque of the assistive device or high precision in the identification of the user intention, the number of classes could therefore be selected appropriately and could be between 13 and 19 classes.
The performance of the classification system in real-time was studied by controlling the WEP by the sEMG signals of the forearm. A control signal was sequentially generated by the system after every 125 ms and the sEMG signals from the data acquisition card was acquired every 125 ms ensuring that the total response time for the system was less than 250 ms. These delays are acceptable for real-time systems as indicated in [29, 30].
Figures 9, 10, 11 and 12 show that the classification system predicts the torque and direction of the user with a good accuracy. The few errors observable in the system also indicate that the misclassified points lie in the adjacent class meaning only the level of torque is incorrectly predicted and not the direction of movement. It is to be noted that the delay in reaching a particular force value for the exoskeleton is due to the response time of the exoskeleton and not to the response time of the classification system.
WEP as an assistive device
This paper explores the possibility of using sEMG signals to control the torque applied by the wrist along with direction of motion. Data was gathered from four forearm muscles during isometric movements of the wrist by using a commercial EMG measurement system and a custom designed rig. sEMG signal rms values, AR model coefficients and waveform length were used to extract features and SVM was used to classify torque of the wrist both into 19 and 13 classes. The average accuracy for 19 classes was about 88% and for 13 classes was 96%. According to the needs of future specific applications, any number of classes in between these two could therefore be potentially suitable. A wrist exoskeleton prototype was developed to study the performance of the real-time system and a proof of concept for the use of WEP as an assistive device was presented. The system was able to respond to user's intention within 250 ms proving that SVM is a suitable technique to be used in real-time sEMG recognition system. The classification system investigated in this study used isometric wrist measurements to simplify the analysis of the investigated problem. Future work will investigate the feasibility of combining force control during dynamic movements.
This work is supported by the Canadian Institutes of Health Research (CIHR), the BC Network for Aging Research (BCNAR), and the Natural Sciences and Engineering Research Council of Canada (NSERC).
- Reaz MBI, Hussain MS, Yasin FM: Techniques of EMG signal analysis: detection, processing, classification and applications. Biol Proced Online 2006, 8: 11–35. 10.1251/bpo115View ArticleGoogle Scholar
- Huang YY, Low KH, Lim HB: Objective and quantitative assessment methodology of hand functions for rehabilitation. Proceedings of the IEEE International Conference on Robotics and Biomemetics: 21–26 February 2009; Bangkok 2009, 846–851.Google Scholar
- Bai O, Nakamura M, Shibasaki H: Compensation of hand movement for patients by assistant force: relationship between human hand movement and robot arm motion. IEEE Trans Neural Sys and Rehab Eng 2001, 9: 302–307. 10.1109/7333.948459View ArticleGoogle Scholar
- Lewis GN, Perreault EJ: An assessment of robot-assisted bimanual movements on upper limb motor coordination following stroke. IEEE Trans Neural Sys and Rehab Eng 2009, 17: 595–604. 10.1109/TNSRE.2009.2029315View ArticleGoogle Scholar
- Lum PS, Burgar CG, Shor PC, Majmundar M, Loos MV: Robot-assisted movement training compared with conventional therapy techniques for the rehabilitation of upper-limb motor function after stroke. Arch Phys Med Rehab 2002, 83: 952–959. 10.1053/apmr.2001.33101View ArticleGoogle Scholar
- Stein J, Narenfran K, Mcbean J, Krebs K, Hughes R: Electromyography-controlled exoskeleton upper-limb-powered orthosis for exercise training after stroke. Am Journal Phys Med Rehab 2007, 86: 255–261. 10.1097/PHM.0b013e3180383cc5View ArticleGoogle Scholar
- The CyberHand project 2007. [http://www.cyberhand.org]
- The SmartHand project 2007. [http://www.elmat.lth.se/~smarthand]
- The iLimb prosthetic hand 2007. [http://www.touchbionics.com]
- Otto Bock SensorHand hand prothesis 2010. [http://www.ottobock.com/cps/rde/xchg/ob_com_en/hs.xsl/3652.html]
- Hudgins B, Parker P, Scott RN: A new strategy for multifunction myoelectric control. IEEE Trans Biomed Eng 1993, 40: 82–94. 10.1109/10.204774View ArticleGoogle Scholar
- Chu JU, Lee YJ: Conjugate-prior-penalized learning of Gaussian mixture models for multifunction myoelectric hand control. IEEE Trans Neural Sys and Rehab Eng 2009, 17: 287–297. 10.1109/TNSRE.2009.2015177View ArticleGoogle Scholar
- Chu JU, Moon I, Lee YJ, Kim SK, Mun MS: A supervised feature-projection-based real-time EMG pattern recognition for multifunction myoelectric hand control. IEEE/ASME Trans Mechatronics 2007, 12: 282–290. 10.1109/TMECH.2007.897262View ArticleGoogle Scholar
- Chu JU, Moon I, Mun MS: A real-time EMG pattern recognition system based on linear-nonlinear feature projection for a multifunction myoelectric hand. IEEE Trans Biomed Eng 2006, 53: 2232–2239. 10.1109/TBME.2006.883695View ArticleGoogle Scholar
- Khezri M, Jahed M: Real-time intelligent pattern recognition algorithm for surface EMG signals. Biomed Eng Online 2007, 6: 45. 10.1186/1475-925X-6-45View ArticleGoogle Scholar
- Englehart K, Hudgins B: A robust, real-time control scheme for multifunction myoelectric control. IEEE Trans Biomed Eng 2003, 50: 848–854. 10.1109/TBME.2003.813539View ArticleGoogle Scholar
- Chan FHY, Yang YS, Lam FK, Zhang YT, Parker PA: Fuzzy EMG classification for prosthesis control. IEEE Trans on Rehab Eng 2000, 8: 305–311. 10.1109/86.867872View ArticleGoogle Scholar
- Karlik B, Tokhi MO, Alci M: A fuzzy clustering neural network architecture for multifunction upper-limb prosthesis. IEEE Trans Biomed Eng 2003, 50: 1255–1261. 10.1109/TBME.2003.818469View ArticleGoogle Scholar
- Park SH, Lee SP: EMG pattern recognition based on artificial intelligence techniques. IEEE Trans on Rehab Eng 1998, 6: 400–405. 10.1109/86.736154View ArticleGoogle Scholar
- Soares A, Andrade A, Lamounier E, Carrijo R: The development of a virtual myoelectric prosthesis controlled by an EMG pattern recognition system based on neural networks. Journal of Intelligent Inf Sys 2003, 21: 127–141. 10.1023/A:1024758415877View ArticleGoogle Scholar
- Castellini C, Smagt PVD: Surface EMG in advanced hand prosthetics. In Biological cybernetics. Springer-Verlag; 2008.Google Scholar
- Liu YH, Huang HP, Weng CH: Recognition of electromyographic signals using cascaded kernel learning machine. IEEE/ASME Trans Mechatronics 2007, 12: 253–264. 10.1109/TMECH.2007.897253View ArticleGoogle Scholar
- Englehart K, Hudgins B, Parker PA, Stevenson M: Classification of the myoelectric signal using time-frequency based representations. Medical Eng & Physics 1999, 21: 431–438.View ArticleGoogle Scholar
- Englehart K, Hudgins B, Parker P: A wavelet-based continuous classification scheme for multifunction myoelectric control. IEEE Trans Biomed Eng 2001, 48: 302–311. 10.1109/10.914793View ArticleGoogle Scholar
- Cipriani C, Antfolk C, Balkenius C, Rosen B, Lundborg G, Carrozza MC, Sebelius F: A novel concept for a prosthetic hand with a bidirectional interface: a feasibility study. IEEE Trans Biomed Eng 2009, 56: 2739–2743. 10.1109/TBME.2009.2031242View ArticleGoogle Scholar
- Chan ADC, Englehart KB: Continuous myoelectric control for powered prosthesis using hidden markov models. IEEE Trans Biomed Eng 2005, 52: 121–124. 10.1109/TBME.2004.836492View ArticleGoogle Scholar
- Bitzer S, Smagt PVD: Learning EMG control of a robotic hand: towards active prosthesis. Proceedings of the IEEE International Conference on Robotics and Automation: 15–19 May 2006; Orlando 2006, 2819–2823. full_textGoogle Scholar
- Yoshikawa M, Mikawa M, Tanaka K: A Myoelectric interface for robotic hand control using support vector machine. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems: 29 October - 2 November 2007; San Diego 2007, 2723–2728.View ArticleGoogle Scholar
- Scott RN: An introduction to myoelectric protheses. UNB Monographs on Myoelectric protheses 1984.Google Scholar
- Farrell TR, Weir RF: The optimal controller delay for myoelectric protheses. IEEE Trans Neural Sys and Rehab Eng 2007, 15: 111–118. 10.1109/TNSRE.2007.891391View ArticleGoogle Scholar
- Bishop CM: Linear models for classification. In Pattern recognition and machine learning. Edited by: Jordan M, Kleinberg J, Scholkopf B. Springer; 2006:179–224.Google Scholar
- Vapnik V: The support vector method of function estimation. In Nonlinear modelling: Advanced black-box techniques. Edited by: Sukens JAK, Vandewalle J. Kluwer academic publishers, Boston; 1998:55–85.View ArticleGoogle Scholar
- Bishop CM: Sparse kernel machines. In Pattern recognition and machine learning. Edited by: Jordan M, Kleinberg J, Scholkopf B. Springer; 2006:325–358.Google Scholar
- Guler I, Ubeyli ED: Multiclass support vector machines for EEG-signals classification. IEEE Trans Inf Tech Biomedicine 2007, 11: 117–126. 10.1109/TITB.2006.879600View ArticleGoogle Scholar
- Khandoker AH, Palaniswami M, Karmakar CK: Support vector machines for automated recognition of obstructive sleep apnea syndrome from ECG recordings. IEEE Trans Inf Tech Biomedicine 2009, 13: 37–48. 10.1109/TITB.2008.2004495View ArticleGoogle Scholar
- Oskoei MA, Hu H: Support vector machine-based classification scheme for myoelectric control applied to upper limb. IEEE Trans Biomed Eng 2008, 55: 1956–1965. 10.1109/TBME.2008.919734View ArticleGoogle Scholar
- Kampouraki A, Manis G, Nikoi C: Heartbeat time series classification with support vector machines. IEEE Trans Inf Tech Biomedicine 2009, 13: 512–518. 10.1109/TITB.2008.2003323View ArticleGoogle Scholar
- Lew HL, TSAI SJ: Pictorial guide to muscles and surface anatomy. In Johnson's practical electromyography. 4th edition. Edited by: Pease WS, Lew HL, Johnson EW. Lippincott Williams & Wilkins; 2007:145–212.Google Scholar
- SENIAM Project [http://www.seniam.org.]
- Luca CJD: Surface electromyography: detection and recording. © 2002 by DelSys Incorporated Google Scholar
- Mital A, Pennathur A: Musculoskeletal overexertion injuries in the United States: mitigating the problem through ergonomics and engineering interventions. Journal of Occupational Rehabilitation 1999, 9: 115–149. 10.1023/A:1021318204926View ArticleGoogle Scholar
- Huang HP, Chen CY: Development of a myoelectric discrimination system for a multi-degree prosthetic hand. Proceedings of the International Conference on Robotics and Automation: May 1999; Detroit 1999, 2392–2397.Google Scholar
- Chang CC, Lin CJ: LIBSVM: a library for support vector machines. 2001. [http://www.csie.ntu.edu.tw/~cjlin/libsvm]Google Scholar
- Hsu CW, Chang CC, Lin CJ: A practical guide to support vector classification. Technical report. Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan; 2003. [http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf]Google Scholar
- Hastie T, Rosset S, Tibshirani R, Zhu J: The Entire Regularization Path for the Support Vector Machine. Journal of Machine Learning Research 2004, 5: 1391–1415.MathSciNetGoogle Scholar
- Sarojini B, Ramaraj N, Nickolas S: Enhancing the Performance of LibSVM Classifier by Kernel F-Score Feature Selection. Proceedings of the Second International Conference on Contemporary Computing: August 17–19; Noida, India 2009, 533–543.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.