International Journal of Information System and Engineering
www.ftms.edu.my/journals/index.php/journals/ijise
Vol. 2 (No.1), April, 2014 Page: 71-78
ISSN: 2289-7615
This work is licensed under a
Creative Commons Attribution 4.0 International License.
Face Recognition System using Template Method
Dr. Srikanth Prabhu
Faculty, Department. of CSE, MIT, Manipal University [email protected]
K. Dashrathraj Shetty
Faculty, Department. of CSE, MIT, Manipal University [email protected]
Manjunath K Vanahalli
Faculty, Department. of CSE, MIT, Manipal University [email protected]
Abstract
Face Recognition is a Biometric Application, which is used for Criminal Identification, Visitor Verification and many other Real Time Identification systems. We use basically two approaches for this system which are namely ‘Template Matching’ and ‘Feature Matching’. The Template Matching approach is independent of the features points, which we have used in this paper. Here we find the convolution values of the features for a test image and all the images in the database. In this work we introduce the novel idea of ‘Energies’. The distance algorithm states that the image in the database having the least distance with the test image in terms of Energies is the identified image.
Keywords: Template, Face, Features, Features Point Field of Research: Image Processing
1. Introduction
Criminal identification systems in modern world works on the phenomena of finger print recognition, face recognition, character recognition and hand writing recognition.
In case of finger print recognition, criminals will not leave any clue as they will be wearing gloves while committing the crime. In face recognition, people (criminals) will be hiding their faces wearing a mask. In character and hand writing recognition, the characters and hand writings of persons will change from time to time. Looking at the conventional biometrics, used now a days, we thought of considering parts of the face as a biometric as there are chances that some part of the face are visible. Going by the literatures, as shown in the next section, we thought of considering the template method for solving the problem. In the template method, the features considered are not the micro features, but macro features, which are not point constricted, but they are area constricted. The feature point method goes by extracting the various features like, eye brows, Lips, eye sockets, nostrils, and then extracting the end points of these features. This type of method, is micro feature based, which needs to be very precise, and if there are false features, it can result in errors. So we thought of going for the template based approach, to overcome the limitations of the feature point based approach.
2. Literature Survey
Researchers all around the world working in the area of biometric are interested in finding out the traits that are unique for an individual. The trait that is unique outside an
individual is face. As pointed out in the earlier section, face is made up of macro features like eye brows, Lips, etc. Researchers have done a considerable amount of research in pattern analysis to extract these features based on color components and intensity parameter [1]. The eye brows and Lips are required for extracting low and high intensity features. Feature points [2] are invariant entities from where distances are calculated. These points are the end points of the macro features, which are also called as micro features. Work has also been done using the template method using Principal Component Analysis or the Genetic algorithms, to extract the features, where the templates, are the different features, which are macro in nature are extracted [3], but doing so, some errors have also been seen, in those features [4]. These errors have been, in the position of the features [5]. So to overcome the errors, we have tried, to extract features, based on segmentation, of the face into three parts. This forms, one of the objectives of the paper. The second objective, after identifying the templates, is to classify the faces, into the corresponding database, using a proper classifier.
3. Objectives
• To identify the macro features, using broad templates, covering a set of features.
• To classify the identified macro features using a classifier.
4. Scope
The comparison of the Energies for the three macro features increases the efficiency of the system as the matching the images get divided three templates, rather than matching being done on the whole image. Hence, comparison with the large data set becomes less resource consuming.
5. Data Collection
The facial images are captured from still photographic cameras, and the aspect ratio of
all images is square, for example 256 * 256. These images are quantified as Energy values (I2), to
extract the features. The images are not of the criminals but of participants who have willingly helped to contribute to the research without any remuneration. The image is shown in Fig 5.1.
The method of extracting the features is shown in Fig 5.2.
Low pass filtering
Medium filtering
Image restoration
Region oriented segmentation
Edge detection operator (Sobel operator)
Fig 5.2: Methodology for extracting features.
The images are first sent through a low pass filter (blurring to remove loss of information), then a median filter (sharpening to remove ambiguities) and later restored. Then region based segmentation is carried out along with edge removal to obtain template based features.
7. Analysis
After applying the Region Based Segmentation [9], we apply the Gabor transform [11], given by
(1)
to get, the different regions of the face, Forehead, Face Snap and the Mouth.
The Scatter Matrix [12] for the above is given by (from Principal Component Analysis [13]) is given by, [21]
(2) The optimized weights are given by [21],
(3)
Using Eq 2 and Eq 3, we get the scatter matrix for the data set, and the weights associated with them.
Then performing Energy analysis for the each of the above three templates for each unit of data as shown in Eq 4 we get [21],
(4)
In Eq 4, the symbols used inside the sets, are the images. Eq 5 gives the risk of misclassification of data [17], in a particular set. This implies [21],
(5)
(6)
Eq 5 and Eq 6 gives the risk as a parameter directly proportional to the loss function, if there are data sets misclassified [21].
(7) (8)
(9)
Eq 7 gives the number of clusters. Eq 8 and Eq 9, gives the net risk value greater than zero [21].
(10) (11) (12)
Eq 11 and Eq 12, gives the risk of misclassification, where the constants are given in Eq 13 and Eq 14 [21].
(13) where, (14) This implies, (15) This implies, (16) where, (17) This implies, (18)
So diagrammatically, we get the sets of proper classification of faces, in Eq 19 [21].
(19)
(20)
(21)
(22) This implies,
(23)
Using the above equations, for Energy Analysis, using Bayesian probabilities, we got an accuracy of 95 %, for a training data set size of 350, and testing data set of size 150, when compared to [8], where features are taken with respect to feature points, and a classification accuracy of 85 % is got.
The results are shown below, with the input and output images as shown in Fig 7.1 and
7.2.
Forehead 5 Forehead 2 Forehead 3
Face snap 4 Face snap 11 Face snap 6
Lips 8.5 Lips 16.6 Lips 8.37
Figure 7.1: Showing the Input Images
Output Images
Forehead 5 Forehead 2 Forehead 3
Face snap 4 Face snap 11 Face snap 5.5
Lips 9 Lips 16.1 Lips 7.87
Figure 7.2: Showing the Matching Images
It is evident from Fig. 7.1 and 7.2 that even though the Person in Fig 7.1 is different from the people in 7.2. The proposed algorithm was able to identify the people who have the similar Fore-head, Face snap and Lips. This would be of great use in Forensic sciences for criminal identification.
The results shown in above section, has given very good classification accuracy of 95 %, when there was a permutation made among the templates, forehead, face snap and Lips.
Further, when compared to feature selection method, false data points, are less in occurrence, in template method, as it deals with the feature as a whole.
These data, used in this paper, are processed sequentially, better classification accuracies can be obtained, in terms of cpu efficiency, and speed up of the processor. Also parallel processing techniques can be used, to remove any ambiguities in the results obtained.
References
Athanasios Nikolaidis, Ioannis Pitas, Facial feature extraction and pose determination Pattern Recognition 33 (2000) 1783–1791.
Pong C. Yuen, J.H. Lai, Face representation using independent component analysis, Pattern Recognition 35 (2002) 1247–1257.
Yingjie Wang, Chin-Seng Chua, Yeong-Khing Ho, Facial feature detection and face recognition from 2D and 3D images, Pattern Recognition Letters 23 (2002) 1191– 1202.
K. Sobottka, I. Pitas, A novel method for automatic face segmentation, facial feature extraction and tracking, Signal Process, Image Commun. 12 (3) (1998) 263– 281.
G. Yang, T.S. Huang, Human face detection in a complex background, Pattern Recognition 27 (!) (1994) 53–63.
C. Kotropoulos, I. Pitas, Rule-based face detection in frontal views, Proceedings of the ICASSP ’97, Munich Germany 1997, pp. 2537–2540.
R. Chellapa, C.L. Wilson, S. Sirohey, Human and machine recognition of faces : a survey, Proc. of the IEEE 83 (5) (1995) 705–740.
R. Brunelli, T. Poggio, Face recognition : features versus templates, IEEE Trans. Pattern Anal. Mach. Intell. 15 (10) (1993) 1042–1052.
J. Illingworth, J. Kittler, The adaptive hough transform, IEEE Trans. Pattern Anal. and Mach. Intell. 9 (5) (1987) 690–698.
R.M. Haralick, L.G. Shapiro, Computer and Robot Vision, Vol. I, Addison-Wesley, Reading MA, 1992.
X. Li, N. Roeder, Face contour extraction from front-view images, Pattern Recognition 28 (8) (1995) 1167–1179.
S. Fischer, B. Duc, Shape normalization for face recognition, AVBPA ’97, Crans-Montana, Switzerland, Lecture Note on Computer Science 1997, pp. 21–26.
M. Kass, A. Witkin, D. Terzopoulos, Snakes : active contour models, Int. J. Comput. Vision 1 (4) (1998) 321–331.
S.R. Gunn, M.S. Nixon, Snake head boundary extraction using global and local energy minimization, Proceedings of the ICPR ’96, Vienna, Austria 1996, pp. 581– 585.
C. Nastar, N. Ayache, Fast segmentation, tracking, and analysis of deformable objects, Proceedings of the ICCV ’93, Berlin, Germany 1993, pp. 275–279.
D.J. Williams, M. Shah, A fast algorithm for active contours and curvature estimation, CVGIP : Image Understanding 55 (1) (1992) 14–26.
K.M. Lam, H. Yan, An improved method for locating and extracting the eye in human face images. Proceedings of the ICPR ’96, Vienna, Austria 1996, pp. 411– 415.
B. Esme, B. Sankur, E. Anarim, Facial feature extraction using genetic algorithms, Proceedings of the ICPR ’96, Vienna, Austria 1996, pp. 1511–1514.
M.J.T. Reinders, P.J.L. van Beek, B. Sankur, J.C.A. van der Lubbe, Facial feature localization and adaptation of a generic face model for model-based coding, Signal Process. Image Commun. 7 (1) (1995) 57–74.
I. Pitas, in : Digital Image Processing Algorithms, Prentice-Hall, UK, 1993. Pattern Classification, Duda and Heart, Second Edition.