Facial gender recognition using Gabor-DCT feature extraction

Journal of Statistics and Management Systems ISSN: 0972-0510 (Print) 2169-0014 (Online) Journal homepage: https://www.tandfonline.com/loi/tsms20 Facial gender recognition using Gabor-DCT feature extraction Shubh Lakshmi Agrwal, Ayushi Jhanwar, Kuldeep Goswami, Sandeep K. Gupta & Vibhor Kant To cite this article: Shubh Lakshmi Agrwal, Ayushi Jhanwar, Kuldeep Goswami, Sandeep K. Gupta & Vibhor Kant (2019) Facial gender recognition using Gabor-DCT feature extraction, Journal of Statistics and Management Systems, 22:4, 719-728, DOI: 10.1080/09720510.2019.1609728 To link to this article: https://doi.org/10.1080/09720510.2019.1609728 Published online: 25 Jun 2019. Submit your article to this journal View Crossmark data Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=tsms20 Journal of Statistics & Management Systems ISSN 0972-0510 (Print), ISSN 2169-0014 (Online) Vol. 22 (2019), No. 4, pp. 719–728 DOI : 10.1080/09720510.2019.1609728 Facial gender recognition using Gabor-DCT feature extraction Shubh Lakshmi Agrwal * Department of Computer Science and Engineering LNM Institute of Information Technology Jaipur 302031 Rajasthan India Ayushi Jhanwar Kuldeep Goswami Department of Computer Science and Engineering Government Women Engineering College Ajmer Ajmer 305002 Rajasthan India Sandeep K. Gupta Jekson Vision Pvt. Ltd. Ahmedabad 380058 Gujarat India Vibhor Kant LNM Institute of Information Technology Jaipur 302031 Rajasthan India Abstract Facial Gender Identification has vast application in human computer interaction, determines customer profile in shopping centers, and restricted permission to enter in prohibited zone, criminal profile analysis. This paper presents a robust process for illumination invariant compact feature extraction using Gabor filter for the automatic *E-mail: [email protected] © 720 S. L. AGRWAL, A. JHANWAR, K. GOSWAMI, S. K. GUPTA AND V. KANT recognition for facial gender identification system. Face has uniqueness in edges and texture pattern for different gender category. Gabor filter can extract edge and textural patterns of faces but generate problem of huge dimensions and redundancy feature coefficients. In order to enhance the efficiency and accuracy of the system, this problem of enormous redundancy as well as dimension can be solved by proposing a new feature namely average-DCT feature reduction technique. Proposed Gabor-DCT has precise, accurate and compact feature pattern as well as early throughput for facial gender identification compared to other state of art method of Gabor filter. Subject Classification: (2010) 68M12 Keywords: Facial gender recognition, Gabor filter, DCT 1. Introduction Faces have wide encoded patterns of different information about human as gender, age, emotion and unique identity. Extraction of information from facial pattern has applicable in various applications in area of authentication of biometric, de-identification, behavioral analysis and human machine interactions [1, 25]. Facial gender identification plays a very important role and has enormous applications in the area of computer vision as determine customer profile in shopping centers, and restricted permission to enter in prohibited zone, criminal profile analysis in real time video streaming. For the recognition of facial genders many research techniques have been suggested by the researchers but scope of results are produced in constrained environment as frontal face images, same scale of face while in real scenario require auto detect face in whole body image and detected face may have rotational angle, different position, size variation and expression variation. Test and validation set may include different age group people. General process of facial gender identification include image acquisition from camera, preprocessing for reducing noise and prepare facial images good for feature extraction, extracting unique feature pattern and classification of gender based on the extracted features set [2]. Static facial gender recognition approach involves evaluating maximum frequency of the evaluated gender in the face images in the process of gender analysis from the face images. The performance of this approach is not good for the commercial applications because it is affected by age as it ignores dynamic gender features in the sequencing of images [3]. The process which makes this technique more efficient in performance for the evaluation of the entire image sequence is the extraction of motion, information of time (changes in facial landmarks and textures etc.) from FACIAL GENDER RECOGNITION 721 each image sequence in dynamic approach [3]. In real world scenario, variation in illumination, position orientation, rotation, scale, obstacle on face provide challenges to existing state art of method in term of accuracy [4]. Gabor filter provides robust result of texture and edge pattern for illumination and scale variation but generate problem of huge dimensions and redundancy of feature coefficients. Redundancy increases confusion and high dimensional feature reduce the performance of system. This research paper include the focus on problem of Gabor filter feature extraction for facial gender identification and proposed the mathematical model as solution and proven with experimentally that proposed solution has optimized the accuracy and solve the problem of high dimensionality and redundancy of Gabor edge features for facial gender recognition. The proposed model reduce the overall size of feature vector of compact energy coefficients for representation of edge, texture features of facial gender class and achieved robust illumination invariant model for facial Gender identification. 2. Literature Review Accuracy of gender recognition depends on preprocessing, extraction of feature, classification. The essential process that can change the accuracy remarkably is the optimization of the extraction of features [5, 26]. Edge patterns and variations in the skin color are the various features for the unique determination of the gender in which texture and edges patterns. Facial expression variation is also a challenge in the Gender classification. Because of the movements in the facial muscles, deformation of the shape of the mouth, eye, skin texture, eyebrows occurs and that’s why gender on the face makes unique shape patterns and textures which confuse feature of facial gender characteristics. Facial gender texture patterns and shapes are consequently taken using feature extraction. Most known features extraction technique for facial gender identification are Gabor filter [6], discrete wavelet transform (DWT), histogram of gradient (HOG) [7], local binary pattern [8], DCT [9], scale invariant feature [10] etc. This enormous dimensions and redundancy of the Gabor filter based facial features evolution methodology was identified by Nguyen et al. [11] and he concluded that local Gabor filters are more advantageous as they take less time for the process of feature extraction. Fusion of different feature extraction method [12] is applied for recognition through merging of different features vectors of Gabor, DCT, DWT and Gaussian but it makes the process of extracting features very complex and time consuming. 722 S. L. AGRWAL, A. JHANWAR, K. GOSWAMI, S. K. GUPTA AND V. KANT Later on, the previously extracted features are optimized using feature reduction. Some of the possible approaches for this are to use concept of Principal Component Analysis (PCA) [13], linear discriminate analysis (LDA) [14]. The results obtained by making use of local spatial features depicts that it extract local maximum pattern with better recognition rate for gender classification. The proposed model Gabor-DCT of feature selection solve problem of existing approach in Gabor features extraction for achieve better performance in accuracy for facial gender identification. 3. Proposed Model The proposed model of Facial gender identification is designed for identification of different textural representation on face of person. These textures must be extracted for classification of facial gender in image or image sequences. 3.1 Face Detection To detect the facial regions using Viola–Jones face detector [15], first of all an image is to be captured by a camera and haar feature is extracted. Multi cascaded adaboost classification discard the non face region in multiple phases of decision and provides high accurate face bounding box in each frame. 3.2 Model of Texture and Edge Feature Evaluation Gabor kernel represented as product of 2D Gaussian kernel and sinusoidal kernel by equation (1) [16]. GK (x , y , λ , θ ) = 1 2πσ xσ y e 2  2 −1  x1 y1 + 2  σ 2 σ 2 x y     2π x1    i    λ  e (1) Here we calculate x1 and y1 using projection angle q as equation 2 [17].  x1 x cos θ + y sin θ =  −x sin θ + y cos θ  y1 = (2) Convolution operation is applied on all pixel of detected face image I(x, y) using above 3x3 Gabor kernel GK(l, q) for generating the edgetexture features of facial gender pattern as equation (3) [17]. FACIAL GENDER RECOGNITION 723 Figure 1 Gabor edge and texture feature evaluated for using convolution operation between facial image and Gabor kernel for various orientations {0, p/4, p/2, 3p/4, p} respectively. Gabor average feature extraction shows cumulative features by one feature matrix of 2 dimensions with reduced redundancy. GFm , n (x , y , ) = I (x , y ) convolution operation GK (λ , θ )m , n (3) Gabor filter create M*Nq different feature matrices. Therefore, it generates 4 dimensionality feature space. Here M represents total number of scale applied in Gabor projection and Nq denotes the number of projection /orientation used in Gabor kernel. The below mentioned meanDCT optimization leads to reduction of dimensionality and redundancy and hence there is visible increment in accuracy of system. 3.3 Average Feature Reduction In the proposed mean feature technique mean value of all same positions Gabor features of 5 Gabor matrix is evaluated as representation in equation 4. This process represents 5 Gabor orientation matrix into single Gabor mean matrix and number of feature are reduced by 1/5. Dimensionality of Gabor matrix is reduced from 15 to 3. It reduced the redundancy of edges as shown in figure 1. Average_G(x, y)m = Matrix_addition(|GF(x, y)|m, 0,|GF(x,y)|m, p/4, |GF(x,y)|m,p/2, |GF(x,y)|m, 3p/4, |GF(x,y)|m, p)/5 (4) 3.4 DCT Feature Extraction Each Gabor-mean feature matrix is transformed into frequency domain using discrete cosine transform DCT [18]. DCT has energy compaction property so small number of DCT low frequency features represents high number of Gabor-mean features. Low frequency region (LL) represents the significant characteristics texture feature of facial gender points as eyes, eyebrows, nose and mouth. The compact energy features from low frequency region of DCT matrix are extracted using 724 S. L. AGRWAL, A. JHANWAR, K. GOSWAMI, S. K. GUPTA AND V. KANT ziz-zag feature selection process and remaining is discarded. DCT feature matrix shows significant characteristics feature of image in top left region and right bottom shows noise and redundant features. So feature selection using ziz-zag process start from top left position and right-bottom positional feature are discarded in Gabor-averageDCT. 3.5 Classification Model The publically available FERET dataset [19] is decomposed into two disjoint sets in ratio of 80/20 in which training part include 80% and testing part include 20%. All collective feature vector with ground truth of training samples is given to support vector machine (SVM) classifier [20] for evaluation of training model. 4. Experiments, Results & Analysis The proposed model is evaluated useing publically available FERET dataset [19] of facial gender identification and ck+ dataset which have challenges as images of different age group, hair style changes, background changes, illumination changes, and different gender. This real life variability makes hard to classify facial gender correctly in these types of images. We have evaluated performance of proposed trained model in to scenario: first, the test results are evaluated with FERET data with Table 1 Results of proposed model of feature extraction for facial gender identification and other state of art results S. No. Model of Feature Extraction Approach MAP (%) 1 Gabor filter + PCA [21] 83.00 2 PCA - SVM [linear kernel] [22] 72.00 3 PCA-SVM[RBF] [22] 77.4 5 Gabor-meanPCA [23] 88.9 6 LBP – PCA [24] 86.5 7 Gabor-PCA- SVM [linear kernel] [24] 84.51 8 Gabor-meanPCA- SVM [linear kernel] [25] 88.47 9 Proposed Model of Gabor-average-DCT feature Evaluation 92 10 Real Time Proposed Gabor-average-DCT feature extraction 88 FACIAL GENDER RECOGNITION 725 Table 2 Accuracy for Each Class of Different Gender Gender Male Female Male Female 90 10 17.90 86.10 five cross validation. Second, the proposed trained model with FERET dataset is evaluated on real time facial gender on web cam and accuracy of facial gender is tested. The performance of proposed model is shown in table 1 with existing state art of results. First experiments with testing of FERET dataset, average accuracy of proposed model is 92% which is far better than other state art of results. Second experiments is achieved 88% accuracy with real life images of webcam tested on proposed trained model with FERET dataset. The accuracy of second experiment is reduced due to more challenges in webcam images as camera noise rotation, variation in background, illumination, pose orientation, distance of face to camera. The increment intensity of these challenges has reduced the accuracy of system in case of real webcam. For FERET dataset, the accuracy of proposed is also evaluated for each gender separately which is shown in table 2. The images of male gender are classified with least accuracy of 90% while images female classes are most accurate in classification. 5. Conclusion In this paper, we focus on the problem of gender detection using Gabor-DCT Feature Extraction. Experimental results analysis show that proposed model of facial gender recognition is far better results in terms of accuracy than existing state of art. The proposed model of facial gender identification reduce the problem of high redundancy, dimensionality and confuse features effectively and optimized manner. References [1] Janu, Neha, Pratistha Mathur, Sandeep Kumar Gupta, and ShubhLakshmiAgrwal. “Performance analysis of frequency domain based feature extraction techniques for facial expression recognition.” In Cloud Computing, Data Science & Engineering-Confluence, 2017 7th International Conference on, pp. 591-594. IEEE, 2017. 726 S. L. AGRWAL, A. JHANWAR, K. GOSWAMI, S. K. GUPTA AND V. KANT [2] Li, B., Lian, X. C., & Lu, B. L., “Gender classification by combining clothing, hair and facial component classifiers”, Neurocomputing, 76(1), 18–27, (2012). [3] BenAbdelkader, C., & Griffin, P. A local region-based approach to gender classification from face images. In Computer vision and pattern recognition-workshops, 2005. CVPR Workshops. IEEE Computer Society Conference on (p. 52). IEEE. (2005, June). [4] Guo, G., Dyer, C. R., Fu, Y., & Huang, T. S. Is gender recognition affected by age? In Computer VisionWorkshops (ICCVWorkshops), 2009 IEEE 12th International Conference on (pp. 2032– 2039). IEEE. (2009, September). [5] Gao,W., & Ai, H. Face gender classification on consumer images in a multiethnic environment. In International Conference on Biometrics (pp. 169–178). Springer Berlin Heidelberg. (2009, June). [6] Haider, K. Z., Nawaz, T., Habib, H. A., Maqsood, M., & Amin, T. U. Gender Classification of Consumer Face Images using Gabor Filters. International Journal of Computer Science and Network Security (IJCSNS), 16(2), 46. (2016). [7] Fu, X., Dai, G.,Wang, C., & Zhang, L. Centralized Gabor gradient histogram for facial gender recognition. In Natural computation (ICNC), 2010 sixth international conference on (Vol. 4, pp. 2070–2074). IEEE. (2010, August). [8] Ojala, T., Pietikinen,M., &Menp, T. Gray scale and rotation invariant texture classification with local binary patterns. In European Conference on Computer Vision (pp. 404–420). Springer Berlin Heidelberg. (2000, June). [9] Smirg, Ondrej, Jan Mikulka, Marcos Faundez-Zanuy, Marco Grassi, andJiri Mekyska. “Gender recognition using PCA and DCT of face images.” In International Work-Conference on Artificial Neural Networks, pp. 220-227. Springer, Berlin, Heidelberg, 2011. [10] Lowe, D. G. Distinctive image features from scale-invariant key points. International journal of computer vision, 60(2), 91–110. (2004). [11] Nguyen, Hieu V., Li Bai, and Linlin Shen. “Local gabor binary pattern whitened pca: A novel approach for face recognition from single image per person.” In International Conference on Biometrics, pp. 269-278. Springer, Berlin, Heidelberg, 2009. FACIAL GENDER RECOGNITION 727 [12] Gupta, Sandeep K., ShubhLakshmiAgrwal, Yogesh K. Meena, and Neeta Nain. “A hybrid method of feature extraction for facial expression recognition.” In Signal-Image Technology and Internet-Based Systems (SITIS), 2011 Seventh International Conference on, pp. 422-425. IEEE, 2011. [13] Rai, P., Khanna, P. A gender classification system robust to occlusion using gabor features based (2D) 2PCA, Journal of Visual Communication and Image Representation, 25, 1118–1129. (2014). [14] Li, Ming, and Baozong Yuan. “2D-LDA: A statistical linear discriminant analysis for image matrix.” Pattern Recognition Letters 26, no. 5 (2005): 527-532. [15] Viola, Paul, and Michael Jones. “Rapid object detection using a boosted cascade of simple features.” In Computer Vision and Pattern Recognition, 2001. CVPR 2001. Proceedings of the 2001 IEEE Computer Society Conference on, vol. 1, pp. I-I. IEEE, 2001. [16] Mehrotra, R., Namuduri, K. R., &Ranganathan, N. Gabor filter-based edge detection. Pattern recognition, 25(12), 1479–1494. (1992). [17] Ignat, A.,&Coman, M. Gender recognition with Gabor filters. In EHealth and Bioengineering Conference (EHB), 2015 (pp. 1–4). IEEE. (2015, November). [18] Ekenel, Hazim Kemal, and Rainer Stiefelhagen. “Analysis of local appearance-based face recognition: Effects of feature selection and feature normalization.” In Computer Vision and Pattern Recognition Workshop, 2006. CVPRW’06. Conference on, pp. 34-34. IEEE, 2006. [19] Phillips, P. Jonathon, Hyeonjoon Moon, Syed A. Rizvi, and Patrick J. Rauss. “The FERET evaluation methodology for face-recognition algorithms.” IEEE Transactions on pattern analysis and machine intelligence 22, no. 10 (2000): 1090-1104. [20] Amari, Shun-ichi, and Si Wu. “Improving support vector machine classifiers by modifying kernel functions.” Neural Networks 12, no. 6 (1999): 783-789. [21] Lemley, Joseph, Sami Abdul-Wahid, DipayanBanik, and RazvanAndonie. “Comparison of Recent Machine Learning Techniques for Gender Recognition from Facial Images.” (2016). [22] Smirg, Ondrej, Jan Mikulka, Marcos Faundez-Zanuy, Marco Grassi, and Jiri Mekyska. “Gender recognition using PCA and DCT of face images.” Advances in Computational Intelligence (2011): 220-227. 728 S. L. AGRWAL, A. JHANWAR, K. GOSWAMI, S. K. GUPTA AND V. KANT [23] Yadav, Pooja, AmarjeetPoonia, Sandeep, and ShubhLakshmiAgrwal. “Performance analysis of Gabor 2D PCA feature extraction for gender identification using face.” In Telecommunication and Networks (TEL-NET), 2017 2nd International Conference on, pp. 1-5. IEEE, 2017. [24] Makinen, Erno, and RoopeRaisamo. “Evaluation of gender classification methods with automatically detected and aligned 2017 2nd International Conference on Telecommunication and Networks (TEL-NET 2017) faces.” IEEE Transactions on Pattern Analysis and Machine Intelligence 30, no. 3 (2008): 541-547. [25] Gupta, Sandeep K., and Neeta Nain. “Gabor Filter meanPCA Feature Extraction for Gender Recognition.” In Proceedings of 2nd International Conference on Computer Vision & Image Processing, pp. 79-88. Springer, Singapore, 2018. [26] S. Ram, S. Gupta, B. Agarwal, “Devanagri Character Recognition Model Using Deep Convolution Neural Network”, In Journal of Statistics and Management Systems, Taylor Francis, 21 (4), pages:593–599, 2018. [27] S. Seth, B. Agarwal, “A hybrid deep learning model for detecting diabetic retinopathy”, In Journal of Statistics and Management Systems, Taylor Francis, 21 (4), pages: 569–574 2018.