Research of Face Recognition with Fisher Linear Discriminant

IOP Conference Series: Materials Science and Engineering You may also like PAPER • OPEN ACCESS Research of Face Recognition with Fisher Linear Discriminant - Ultra-High Frequency Partial Discharge Signal Recognition in GIS based on Fisher Linear Discriminant Theory Pei Xudong, Pei Xubing and Xu Haiqi To cite this article: R Rahim et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 300 012037 - The Exploring of Student’s School Majoring Data at Madrasah Aliyah Negeri in Samarinda Using Linear Discriminant Analysis Models N A Rizki, Wasono and Y N Nasution View the article online for updates and enhancements. - Vowel Recognition Based on Face Images Using Fisher Linear Discriminant Analysis Lina and Desi Arisandi This content was downloaded from IP address 34.228.12.132 on 16/01/2023 at 06:00 4th International Conference on Operational Research (InteriOR) IOP Publishing IOP Conf. Series: Materials Science and Engineering 300 (2018) 012037 doi:10.1088/1757-899X/300/1/012037 1234567890‘’“” Research of Face Recognition with Fisher Linear Discriminant R Rahim1*, T Afriliansyah2, H Winata3, D Nofriansyah4, Ratnadewi5 and S Aryza6 1 Department of Informatics Engineering, Institut Teknologi Medan, Medan, Indonesia 2 STKIP Bumi Persada Lhokseumawe, Jalan Medan Banda Aceh No. 59, Lhokseumawe, Indonesia 3 Department of System Computer, STMIK Triguna Dharma, Jl. Jenderal Abdul Haris Nasution No.73, Medan 20219, Indonesia 4 Department of Information System, STMIK Triguna Dharma, Jl. Jenderal Abdul Haris Nasution No.73, Medan 20219, Indonesia 5 Electrical Engineering, Maranatha Christian University, Jalan Prof. Drg. Surya Sumantri No.65, Bandung 40164, Indonesia 6 Faculty Engineering, Universitas Pembangunan Panca Budi, Jl. Jend Gatot Subroto KM. 4.5, Medan, 20122, Indonesia 1,6 School of Computer and Communication Engineering, Universiti Malaysia Perlis, Perlis, Malaysia *[email protected] Abstract. Face identification systems are developing rapidly, and these developments drive the advancement of biometric-based identification systems that have high accuracy. However, to develop a good face recognition system and to have high accuracy is something that's hard to find. Human faces have diverse expressions and attribute changes such as eyeglasses, mustache, beard and others. Fisher Linear Discriminant (FLD) is a class-specific method that distinguishes facial image images into classes and also creates distance between classes and intra classes so as to produce better classification. 1. Introduction Technological developments are growing rapidly today especially in the field of information technology [1], various applications have been designed to seek information about one's self-data, one of which is the application of self-knowledge [2]. Facial recognition is one of the pattern recognition approaches for personal identification purposes in addition to other biometric approaches such as fingerprint recognition, signature, and retina of the eye and so on [2]. The recognition of the face image is related to an object that is never the same, because of the parts that can change [2] [3] [4]. These changes can be caused by facial expressions, shooting angles, or changes in accessories on the face. In this experiment, the same object with some of the differences must be identified as a single object [5]. Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Published under licence by IOP Publishing Ltd 1 4th International Conference on Operational Research (InteriOR) IOP Publishing IOP Conf. Series: Materials Science and Engineering 300 (2018) 012037 doi:10.1088/1757-899X/300/1/012037 1234567890‘’“” Generally, facial recognition has several stages to identify the face of face detection, face alignment, feature extraction and also face matching [6]. Many methods are used in facial recognition applications, one of which is the Fisher Linear Discriminant method which is the basis of the Fisher face and the development of Principal Component Analysis (PCA) [2] [5] [7]. The Fisher Linear Discriminant (FLD) method is used to find the average face value of each class and to calculate the distribution matrix in the class and the spreading matrix between the classes and to find the optimal projection value of the image to obtain the face weights [5] [8]. The Fisher Linear Discriminant Method (FLD) is the most widely used method for pattern recognition. Fisher Linear Discriminant (FLD) is an example of a class-specific method, as it seeks to form the scatter between classes and intra classes so as to produce a better classification [9]. This paper conducted experiments to test the resemblance of the face with Fisher Linear Discriminant algorithm, the testing process is done gradually by showing the workings of Fisher Linear Discriminant in examining the similarity of the face and compare the image in the test so that the percentage of similarity of the face is known. 2. Fisher Linear Discriminant Fisher Linear Discriminant (FLD), also known as Linear Discriminant Analysis (LDA) was discovered by Robert Fisher in 1936 for taxonomic classification and became one of the most widely used techniques in pattern recognition [3] [4] [6] [9]. FLD is an example of a class-specific method, and it is a well-known method for classifying class into class-and-class matrices of classes that could be taken for classification, as it seeks to form inter-class and intra-class scatterings so as to produce a ranking better [6] [9] [10]. Following steps do the calculation of Fisher Linear Discriminant (FLD) method: a. Specify the number of classes and members of each class. b. Establish the average value of the training set (m) and the average face value of each class (average class face, mi), The formula as below = ∑ (1) The formula calculation looks for the average value of each class: = ∑ , ′ ∈ ′ (2) c. Perform an inter-class matrix of inter-class (between-class scatter matrix, Sb) and in-class scattermatrix, SW matrices. By calculating the formula: ∑ ∑ , ∈ ( − )( − ) (3) = ∑ ( − )( − ) d. Find eigenvalue or WFLD projection matrix with formula calculation: &' . )* ! = " #$% ! = & = ( (4) (5) e. Calculate the value of eigenvector (w) by the formula: w = )* ( − + ) (6) f. Calculate the total scatter matrix to obtain the PCA projection matrix = ∑ ( − )( − ), (7) g. Calculate the optimal projection matrix -./ (8) -./ = ! 0 1 h. Calculate the weight value of each of the face images with the optimal projection matrix -./ with the calculation of the formula. 2 = (9) -./ Before performing face recognition with Euclidean Distance similarity calculation [11] [12], the facial image to be tested with the average value of training set with the formula calculation: ϕ 4.5/ = 4.5/ – m (10) 2 4th International Conference on Operational Research (InteriOR) IOP Publishing IOP Conf. Series: Materials Science and Engineering 300 (2018) 012037 doi:10.1088/1757-899X/300/1/012037 1234567890‘’“” After the normalization value of the test face image is obtained, then the further project is examined into the optimal projection matrix. 2/67/ = -./ ϕ 4.5/ (11) 3. Result and Discussion The experiments were conducted by examining 3 (three) face images that have similarities, of the three face images compared and know which image has the highest similarity, the testing process is done by using Fisher Linear Discriminant algorithm with the stages that have been determined, for more details can be considered in the following process: a. Grayscale Process and Threshold Image Training The face image is changed from the RGB shape into the grayscale form and then turn into a black and white image by doing the threshold, and the result is as follows: Figure 1. RGB Image Change to Grayscale and Threshold Once the training image is formed in the threshold, three images can be obtained with a size of 3 x 3 pixels as can be seen in Figure below 214 106 190 8236 234 223A 253 212 250 139 0 127 8196 174 209A 117 153 178 180 23 151 8204 173 173A 229 249 225 Figure 2. Image Training Results Threshold b. Fisher Linear Discriminant Calculation The use of the formula from steps 1-8 will be tested by the formula already described previously with the pixel value contained in FIG. 2, the result of the testing process using the Fisher Linear Discriminant method and the Euclidean distance to recognize the face can be seen in the following tables: 3 4th International Conference on Operational Research (InteriOR) IOP Publishing IOP Conf. Series: Materials Science and Engineering 300 (2018) 012037 doi:10.1088/1757-899X/300/1/012037 1234567890‘’“” Table 1. Weight Training Image Value Face of 1 2 j Fj,1 Fj,2 Fj,3 Fj,4 Fj,5 1 2 3 -185.072 -104.195 -135.977 -189.196 -96.621 -130.039 -172.310 -83.608 -114.449 -238.204 -201.076 -191.628 -259.480 -214.892 -206.433 Fj,6 -242.412 -199.871 -192.679 Fj,7 -249.451 -151.034 -239.666 Fj,8 -266.149 -164.574 -263.224 Fj,1 -247.350 -154.433 -247.824 Fj,6 -243.412 -199.871 -192.679 1 2 Table 2. Different Squares of Each Image Training of Imagery Test j (Fj,1 – Fuji,1)2 (Fj,2 – Fuji,2)2 (Fj,3 – Fuji,3)2 1 ((-185.072)-(20.32))2 2 2 ((-104.195)-(20.32)) ((-96.621)-(35.52)) 3 j 1 2 3 ((-135.997)-(20.32))2 (Fj,6 – Fuji,6)2 ((-243.412)-(21.86))2 ((-199.871)-(21.86))2 ((-192.679)-(21.86))2 ((-130.039)-(35.52))2 (Fj,7 – Fuji,7)2 ((-249.451)-(16.39))2 ((-151.034)-(16.39))2 ((-239.666)-(16.39))2 ((-189.196)-(35.52))2 2 (Fj,4 – Fuji,4)2 ((-172.310)-(13.98))2 (Fj,5 – Fuji,5)2 ((-238.204)-(13.79))2 ((-259.480)-(34.76))2 ((-83.608)-(13.98)) 2 ((-201.076)-(13.79)) ((-214.892)-(34.76))2 ((-114.449)-(13.98))2 (Fj,8 – Fuji,8)2 ((-266.149)-(27.43))2 ((-164.574)-(27.43))2 ((-263.224)-(27.43))2 ((-191.628)-(13.79))2 (Fj,9 – Fuji,9)2 ((-248.350)-(9.86))2 ((-154.433)-(9.86))2 ((-247.824)-(9.86))2 ((-206.433)-(34.76))2 (Fj,6 – Fuji,6)2 ((-243.412)-(21.86))2 ((-199.871)-(21.86))2 ((-192.679)-(21.86))2 2 Table 3. Sum of Squares of Each Image of Training on Test Image j 1 2 3 j 1 2 3 (Fj,1 – Fuji,1)2 42187.51 15505.57 24430.75 (Fj,6 – Fuji,6)2 65677.84 49166.80 49166,80 46029.08 (Fj,2 – Fuji,2)2 49161.26 17463.19 27412.23 (Fj,7 – Fuji,7)2 70674.14 28032.50 28032,50 65567.28 (Fj,3 – Fuji,3)2 34705.86 9524.41 16495.31 (Fj,8 – Fuji,8)2 86191.62 36867.49 36867,49 84482.71 (Fj,4 – Fuji,4)2 63503.49 46169.54 42198.60 (Fj,9 – Fuji,9)2 66158.21 26992.97 26992,97 66402.28 (Fj,5 – Fuji,5)2 86852.47 62330.6 58178.40 Sum Result (Fj,i – Fuji,i)2 565112.4 292053.08 292053,08 431196.64 Table 4. Final Result Calculation of Euclidean Distance Similarity Sum Results dv=∑m j=1DFtrain of j,w - Ftest,w D (Fj,i – Ftest,i)2 1 565112.4 751.7396 2 292053.08 540.4193 3 431196.64 656.6556 Face - j 1 2 Result Minimal Table 4 generates a minimum value of 540.41935. The value lies on the 2nd face so that the face recognition tests have similarities on the 2nd object. The output of this facial recognition process is a recognizable facial image. 4. Conclusion Based on the discussions and evaluations that have been done on facial recognition application using Fisher Linear Discriminant (FLD) method, it is concluded that the application of Fisher Linear Discriminant (FLD) method in face recognition has high accuracy because this method divides class specifically and face recognition application with Fisher Linear Discriminant (FLD) method is able to recognize faces with various expressions are also able to recognize the face with attribute changes such as glasses, whiskers, and others. 4 4th International Conference on Operational Research (InteriOR) IOP Publishing IOP Conf. Series: Materials Science and Engineering 300 (2018) 012037 doi:10.1088/1757-899X/300/1/012037 1234567890‘’“” References [1] R. Rahim, A. S. Ahmar, A. P. Ardyanti and D. Nofriansyah, "Visual Approach of Searching Process using Boyer-Moore Algorithm," in International Conference on Information and Communication Technology 2017, Medan, 2017. [2] S. Thakur, J. K. Sing, D. K. Basu and M. Nasipuri, "Face Recognition Using Fisher Linear Discriminant Analysis and Support Vector Machine," Springer : International Conference on Contemporary Computing, pp. 318-326, 2009. [3] J. Pan, X. S. Wang and Y. H. Cheng, "Single-Sample Face Recognition Based on LPP Feature Transfer," IEEE Access, vol. 4, pp. 2873 - 2884, 2016. [4] H. Shin, "An extension of Fisher's discriminant analysis for stochastic processes," Elsevier : Journal of Multivariate Analysis, vol. 99, no. 6, pp. 1191-1216, 2008. [5] B. Zou, L. Li, Z. Xu, T. Luo and Y. Y. Tang, "Generalization Performance of Fisher Linear Discriminant Based on Markov Sampling," IEEE Transactions on Neural Networks and Learning Systems, vol. 24, no. 2, pp. 288 - 300, 2013. [6] Y. Li, W. Yuan, H. Sang and X. Li, "Combination recognition of face and ear based on two-dimensional fisher linear discriminant," in 4th IEEE International Conference on Software Engineering and Service Science (ICSESS), Beijing, China , 2013. [7] Y. Xu, Z. Li, J. Yang and D. Zhang, "A Survey of Dictionary Learning Algorithms for Face Recognition," IEEE Access , pp. 1-12, 2017. [8] Y. Zhang and Y. Wo, "A fusion iris feature extraction method based on fisher linear discriminant," in International Conference on Machine Learning and Cybernetics (ICMLC), Tianjin, China , 2014. [9] F. Z. Okwonu and A. R. Othman, "A Model Classification Technique for Linear Discriminant Analysis for Two Groups," International Journal of Computer Science , vol. 9, no. 3, pp. 125-128, 2012. [10] B. Sunitha and M. Seetha, "ENHANCING THE PERFORMANCE OF SVM CLASSIFICATION BASED ON FISHER LINEAR DISCRIMINANT," International Journal of Research in Engineering and Technology, vol. 3, no. 5, pp. 48-54, 2014. [11] R. Rahim, S. Nurarif, M. Ramadhan, S. Aisyah and W. Purba, "Comparison Searching Process of Linear, Binary and Interpolation Algorithm," in International Conference on Information and Communication Technology 2017, Medan, 2017. [12] R. Rahim, M. Dahria, M. Syahril and B. Anwar, "Combination of the Blowfishand Lempel-Ziv-Welch algorithms for text compression," World Transactions on Engineering and Technology Education, vol. 15, no. 3, pp. 292-297, 2017. 5