Computational Complexity Analysis

2 A NTENNA D ESIGN

3.4 Computational Complexity Analysis

The first three steps of the proposed system can be performed using only simple image processing tools such as image resizing, filtering and point selection. The last step can be done using only simple and tiny arithmetic. Both are fully computationally non-intensive operations.

The proposed system is examined on a heterogeneous dataset collected for real world experiment environment.

It can be observed evidently that the performance of the proposed system is better than existing systems in broad perspectives.

4 C

ONCLUSION

This system is based on new features selection for hand geometry based personal verification systems. This systems is very much user friendly and convenient to implement. The system is peg free and images can be captured by a normal scanner. Or- ganizer need not to manage a high image captured instrument or pegged scanner so it can be implemented in where any time.

User can place his/her hand in any orientation less than 45 degree along vertical axis. Generally a large number of feature decrease the performance of computation here only one set of feature vector (nine features) are used which improves the computational efficiency. One special contribution of this system is that it can detect actual valley points although there is a small gap between tips of two fingers. This system runs on hand

with nailed finger accurately. The remarkable achievement obtained from the proposed method is the result of verification, which is best among the prevailing techniques of hand geometry based verification system. The system showed promising results with accuracy around 99.98%. The FRR is found to be close to 0.02 and the FAR to be around 0.02.

The proposed approach utilizes primarily the geometry of the hand and work on colored images. If a grayscale image is uti- lized for the system, then databases search will take a short time and decreases computational time more. If the system works properly when a user placed his/her hand in any angle, it will be more users friendly. The use of neural network based classifier trained on a larger database may result in further improvement of the system accuracy.

REFERENCES

[1] A. Jain, L. Hong, S. Pankanti, and R. Bolle, “Online identity- authentication system using finger-prints," Proceedings of IEEE, vol. 85, pp. 1365-1388, September 1997.

[2] J. Ashbourn, “Biometrics: Advanced Identity Verification”, Springer-Verlag, New York, 2000.

[3] R. Sanchez-Reillo, “Hand geometry pattern recognition through Gaussian mixture modeling”, in 15th International Conference on Pattern Recognition, vol.2, Sep, 2000. pp. 937-940.

R.H.M.A. KABIR ET AL.: A SIMPLE APPROACH TO RECOGNIZE A PERSON USING HAND GEOMETRY 25

[4] A.K. Jain and N. Duta, “Deformable matching of hand shapes for verification”, IEEE International Conference on Image Processing, Oct, 1999. Pp. 857-861.

[5] J. R. Young and H. W. Hammon, “Automatic Palmprint Ve- ri_cation Study", Rome Air Devel- opment Center, RADC-TR- 81-161 Final Techni- cal Report, June1981.

[6] R. Zunkel, “Hand Geometry Based Authentication" in “Biome- trics: Personal Identification in Networked Society", A. Jain, R.

Bolle, and S. Pankanti (Eds.), Kluwer Academic Publishers, 1998.

[7] A. K. Jain, A. Ross, and S. Prabhakar, Feb. 2004, “An introduc- tion to biometric recognition,” IEEE Trans. Circuits Syst. Video Technol., vol. 14, no. 1, pp. 4–20.

[8] R. L. Zunkel, 1999, “Hand geometry based verification,” in Biometrics, A. Jain, R. Bolle, and S. Pankanti, Eds. Norwell, MA: Kluwer, pp. 87–101.

[9] R. Sanchez-Reillo, C. Sanchez-Avila, and A. Gonzalez-Marcos, Oct. 2000, “Biometric identification through hand geometry measurements,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 10, pp. 1168–1171.

[10] A. K. Jain, A. Ross, and S. Pankanti, Mar. 1999, “A prototype hand geometry based verification system,” Proc. 2nd Int. Conf.

Audio- and Video-Based Biometric Person Authentication, pp. 166–

171.

[11] C. Öden, A. Erçil, and B. Büke, 2003, “Combining implicit po- lynomials and geometric features for hand recognition,” Pat- tern Recognit. Lett., vol. 24, pp. 2145–2152.

[12] Y. A. Kumar, D. C.M.Wong, H. C. Shen, and A. K. Jain, Jun. 9–

11, 2003, “Personal verification using palmprint and hand geometry biometric,” Proc. 4th Int. Conf. Audio Video-Based Biometric Person Authentication, Guildford, U.K., pp. 668–678.

[13] Alexandra L.N. Wong and Pengcheng Shi “Peg-Free Hand Geometry Recognition Using Hierarchical Geometry and Shape Matching”, Department of Electronic and Electrical En- gineering, Hong Kong University of Science and Technology.

[14] Ying-Han Pang, Andrew, David and Hiew Fu San, February 2-6, 2003, “Palmprint Verification with Moments”, Journal of WSCG, vol.12, no.1-3, ISSN 1213-6972, 2004.

R.H.M. Alaol Kabir received BSc degree in Com- puter Science and Engineering from University of Chittagong, Bangladesh, in 2008, MS degree in Information Technology from University of Dhaka, Bangladesh, in 2010. He is a Lecturer of Darul Ihsan University, Dhaka, Bangladesh. His research interest includes Biometrics, Image Processing, Biomedical Engineering, and Wireless Communication Networks.

Md. Atikur Rahman received BSc degree in Computer Science and Engineering from Universi- ty of Chittagong, Bangladesh, in 2008. He is a Lecturer of International Islamic University Chitta- gong, Bangladesh. His research interest includes Biometrics, Image Processing, Biomedical Engi- neering, and Microprocessor Architechture.

Mohammad Ahsanul Haque received BSc degree in Computer Science and Engineering from University of Chittagong, Bangladesh, in 2008.

Currently, he is pursuing his MS degree in Com- puter and Information Technology in the University of Ulsan, South Korea. He is a Lecturer of Interna- tional Islamic University Chittagong, Bangladesh.

His research interest includes Biometrics, Image Processing, Embedded Ubiquitus Computing System Design, Multimedia Processing, and Mul- ticore System Design.

Mohammad Osiur Rahman received B.Sc.

Engg. degree in Computer Science and Engi- neering from Shahjalal University of Science &

Technology, Bangladesh in1997, and M.Sc.

Engg. degree in Information & Communication Technology from Bangladesh University of En- gineering Technology in 2005. Currently he is pursuing his PhD in the department of Electrical , Electronic and Systems Engineering, Universi- ty Kebangsaan Malaysia, Malaysia. He is an Assistant Professor in the CSE department of the University of Chittagong, Bangladesh. His research interest includes Biometrics, Image Processing, DNA Computing, and Pattern Recognition.

M.H.M. Imrul Kabir received BSc degree in Applied Statistics from University of Dhaka, Bangladesh, in 2008, MS degree in Applied Statistics from University of Dhaka, 2010. His research interest includes Statistical Machine Translation, Biometrics,and Image Processing,.

ULAB JOURNAL OF SCIENCE AND ENGINEERING VOL. 1, NO. 1, NOVEMBER 2010 (ISSN: 2079-4398) 26

Human Emotion Recognition Using PCA, ICA and NMF

Paresh Chandra Barman, Chandra Shekhar Dhir, Soo-Young Lee

Abstract—Recognizing human facial expression and emotion by computer is an interesting and challenging problem. In this paper, Principal component analysis (PCA), independent component analysis (ICA) and Non-negative Matrix Factorization (NMF) are exploited for feature extraction of face images. The features are low-dimensional representation of the original multivariate high dimensional data with minimal loss in data representation [1]. In addition, the features are also required to give good class discrimination for recognition experiments. Feature selection based on information gain criterion has been studied for finding efficient features to improve the classification performance of face recognition tasks [2]. This work presents a detailed study on the application of information gain for efficient feature selection and is compared with Fisher criterion. Individual, emotion recognition experiments using face images of Korean nationals are performed to compare the two feature selection criteria (Information gain and Fisher criterion). The face images of Korean nationals are obtained from the Postech Faces ’01 (PF01) database [3].

Keywords—Human emotion recognition, NMF, PCA, ICA.

1 I

NTRODUCTION

UMAN-computer intelligent interaction (HCII) is an emerging field of science aimed at providing natu- ral ways for humans to use computers as aids. It is argued that for the computer to be able to interact with humans, it needs to have the communication skills of humans. One of these skills is the ability to understand the emotional state of the person. The most expressive way humans display emotions is through facial expressions. In recent years there has been a growing interest in improving all aspects of the interaction between humans and computers. Faces are much more than keys to individual identity. Human beings possess and express emotions in day to day interactions with others.

The problem of handling facial images for recognition is also due to the large pixel size and demands on computational resources. Therefore, several researchers have exploited linear low-dimensional representation of images using orthogonal basis by Principal component analysis (PCA) and independent features using Independent component analysis (ICA). ICA is a multivariate approach of data representation with statistically independent features [5]. It has also found wide applications in the field on blind source separation. In [15] Cauchy Naive Bayes Classifier has been used for emotion recognition.

Face recognition using eigenfaces was one of the prelimi- nary works in the field of face recognition using low dimensional features [6]. While ICA and NMF give more local representation of data, PCA features are global. Al-

though, feature extraction methods helps in low- dimensional representation of multivariate data by using unsupervised methods, all the feature may not be important for classification. Hence, the features extracted give a good data representation and the next task is to select proper features for good classification performance. Va- riance of features and Fisher criterion (between class variance over within class variance) has been widely used for selection of features for classification [7]. Variance of feature gives an estimate of the power of the features that can be considered important. Recently, feature selection criterion based on information gain to face images is studied for efficient feature selection and improving classification performance [2].

A comparative study on the performances of the features will be done on feature selection criterion based on information gain to facial images is studied for efficient feature extraction and improving classification performance. Information gain criterion is extensively studied in the field of text categorization [8]. The motivation to apply information gain was to maximize the information between the class and the given features. Since, the ICA features are independent we can get a score value for each feature based on information gain and a proper number of features can be selected. However, in case of PCA and NMF features which are dependent this criterion in its crude form cannot be applied. The information gain criterion presented in the paper does not consider the dependency among features which may be present in case of PCA and NMF features.

In section II, feature extraction using PCA, ICA and NMF are discussed. Section III concentrates on feature selection using Fisher criterion and proposed information gain for application to face recognition. Experimental setup is presented in section IV followed by results in section V. Sec-

H

————————————————

 Paresh Chandra Barman is with the Department of Information and Com- munication Engineering, Kushtia, Bangladesh. E-mail: [email protected].

 Chandra Shekhar is with the Department of Bio and Brain Engineering, KAIST, Republic of Korea.

 Soo-Young Lee is with the Department of Electrical Engineerig, KAIST, Republic of Korea.

Manuscript received on 31 July 2010 and accepted for publication on 27 August 2010.

PARESH C. BARMAN ET AL.: HUMAN EMOTION RECOGNITION USING PCA, ICA AND NMF 27

tion V1 discusses the experimental results followed by conclusions.

A brief overview on the two dataset used is given below:

In this paper we utilize the publicly available Postech Faces ’01 (PF01) *3+ face database. PF01 has 56 males’ faces with 4 emotions of each, and the resolution is 150150.

The four different emotion categories are: Smile, Surprised, Sad and Closed eyes as shown in figure1. There are 168 training samples per class, 56 test samples per class. We used 4 fold cross-validations for recognition process.

Figure 1: Example of 4 facial emotion expressions.

2 F

EATURE

E

XTRACTION

M

ETHODS

Classification of emotion images with high resolution is a difficult problem and is computationally demanding if no pre-processing is done on the raw images. In an at- tempt to reduce the computational burden several efficient feature extraction methods have been studied. PCA is a well known method in image processing and has been widely used to extract meaningful features. Working on same philosophy, ICA and NMF are also used for representing low dimension feature space.

Dalam dokumen ULAB JOURNAL OF SCIENCE AND ENGINEERING (Halaman 31-34)