Multilayer perceptrons with one input, one or more hidden layers and one output .... database http://cswww.essex.ac.uk/mv/allfaces/index.html. REFERENCES.
1
Face Recognition using morphological method Dr. Ritu Tiwari, Dr. Anupam Shukla, Chandra Prakash, Dhirender Sharma , Rishi Kumar, Sourabh Sharma Department of Information Communication and Technology, Indian Institute of Information Technology and Management Gwalior, India
Abstract— In this paper we discussed and implemented Morphological method for face recognition using fiducial points. A new technique for extracting facial features is suggested here. This method is independent of the face expressions. In recognition process, these fiducial point are fed as inputs to a Back propagation neural network for learning and identifying a person. So with the help of this technique, recognition is applied under various facial expressions. This algorithm is tested on Grimice Database and compared with existing feature extraction techniques PCA and R-LDA. Morphological method is able to give recognition results better than PCA extracted features.
Index Terms – Edge detection; Face Recognition; Facial features; Feature extraction; Morphological Method Image processing; fiducial point.
I. INTRODUCTION ECURITY is the one of the main concern in today’s world. Whether it is the field of telecommunication, information, network, data security, airport or home security, national security or human security, there are various technique for the security. Biometric is one of the mode of the security. A biometrics is, “Automated methods of recognizing an individual based on their unique physical or behavioral characteristics.”[1]. Biometrics consists of two classes: Physiological and Behavioral. Physiological traits include face, fingerprint, iris, hand and DNA. Signature and voice are the part of Behavioral class.
S
In the fear of the terrorism the face recognition techniques are used at airport, seaport in the case of national security. But the methods used at these places are conventional biometric methods and thus static as it is mainly based on cross matching the face the traveler with that in the international passport and fingerprints. At present, identification of the suspects with the old technique is not reliable in case of cosmetic or plastic surgery modification of their faces. Although a new technique in which database of biometric information including faces and fingerprints are used to identify a person. Identification and Verification are the two main function of the Recognition system. Verification is Authenticating identity of an individual by comparing a presented characteristic to a pre-enrolled characteristic while Identification is determining the identity of an individual by comparing a presented characteristic to all pre-enrolled in the database. In simple words “Are you who you say you are ?” is Verification and “Who are you?” is Identification.
For the Researchers Face Recognition is among the trides work. It is all because the human face is very robust in nature; in fact, a person’s face can change very much during short periods of time (from one day to another) and because of long periods of time (a difference of months or years). One problem of face recognition is the fact that different faces could seem very similar; therefore, a discrimination task is needed. On the other hand, when we analyze the same face, many characteristics may have changed. These changes might be because of changes in the different parameters. The parameters are: illumination, variability in facial expressions, the presence of accessories (glasses, beards, etc); poses, age, finally background [2],[3],[4].We can divide face recognition techniques into two big groups, the applications that required face identification and the ones that need face verification. The difference is that the first one uses a face to match with other one on a database; on the other hand, the verification technique tries to verify a human face from a given sample of that face [5] . Feature extraction methods can be distinguished into three types: (1) generic methods is based on the analysis of edges, lines, and curves ; (2) feature-template-based methods is based on the detection of the facial features such as eyes; (3) structural matching methods that take into consideration geometrical constraints on the features [1],[6],[7]. The technique we proposed here is independent of the aging factor, illumination and presence of accessories (glasses, beards, etc). Here in this technique we are considering the fiducial points. The points are the distance between eyes; eye and mouth; eyes and nose. The distance between these facial points never changes. Then we implement the Back Propagation Neural Network (BP-NN) to the system Details of this system are described in the remainder of this paper. The paper is organized as follows: Section (II) covers our Methodology of the system, section (III) we present a method for face recognition based on fiducial points and BPA- NNs. In Section IV, experimental results of evaluating the developed techniques are presented. Finally, conclusions with limitation and future work are summarized in Section V. II. METHODOLOGY OF THE SYSTEMS In this section we introduce our technique to find the facial feature in the still colored image. Our methodology of recognition of faces involves four phases: Preprocessing, Segmentation of faces it include face detection from scenes,
2
feature extraction from the face regions and finally recognition of the face. A. Preprocessing In this section we discuss the various techniques we had used before finding the facial features in the image. It is also known as normalization process. The intensity of light in the image is not unique so our first step is to make the image equally enrich. First we take the input image as color image. [Figure 1(1)] Then we converted to gray scale. [Figure 1(2)] Then on the converted images segmentation techniques are used. B. Segmentation Segmentation is one of the very first steps in automatic face recognition systems. The goal of segmentation is making the image more analyzable. In simple words object detection is considered as segmentation. Up to the mid-1990s, Segmentation’s main focused was on single-face segmentation from a simple or complex background. Significant advances have been made in recent years in achieving automatic face detection under various conditions [8],[ 9].
Figure1 : Step to get segmented image
There is a difference in the object to be segmented and the background image in case of contrast. By calculating changes in contrast within the image we can calculate the gradient of an image. After calculating gradient image, edge and Sobel operators are used to calculate the threshold value, which in turn give a binary gradient image. [Figure 1(3)]. The processed binary gradient mask images still shows lines of high contrast in the image by using linear structuring elements i.e dilating of the binary gradient image, these linear gaps can be removed. [Figure 1(4)] This is very similar to the human perceive the face of other human being. Then region filling to get binary image with filed hole [figure 1(5)]; Extraction of 8-Connected set of pixels components to suppress light structures connected to image border [figure 1(6)]; filtering; thinning and Pruning [10] are implementation result in segmented image [figure 1(7)]. The segmented image is then superimposed with the initial gray image [figure1 (8)]. C. Feature Extraction
3
The extracted features like eyebrows, eye, nose and mouth are now in enhanced form. Feature Extraction algorithm includes: a) Selection of the more accurate features b) Determination of Normal Center of Gravity (NCG) The segmented image is processed with the proposed algorithm of finding more accurate features. The algorithm results in removal of small objects and results in morphologically open binary image. Once the features are identified, the algorithm determines the Normal Center of Gravity (NCG) or Intensity-weighted centroids of the each extracted features. The NCG is shown in the Figure [2] with blue stars. [10].
III. FACE RECOGNITION USING BP-NN In recognition process, these Euclidean distances are fed as inputs to a Back propagation neural network for learning and identifying a person. The experiments were performed with ten distinct subjects and with Grimace Face database. A. Back Propagation-Multi-Layer Neural Networks Structure: The Back Propagation algorithm is a supervised learning method and looks for the minimal of error function in weight space using the method of gradient descent and hence continuity and differentiability of error function is mandatory. The combination of weight which minimizes the error function is the solution of the learning problem. The calculated error is back propagated from one layer to the previous one, and is used to adjust the weights between connecting layers. Training stops when error becomes acceptable, or after a predetermined number of iterations. After training, the modified interconnection weights form a sort of internal representation that enables the ANN to generate desired outputs when given the training inputs – or even new inputs that are similar to training inputs. Back propagation usually allows quick convergence on satisfactory local minima for error in the kind of networks to which it is suited.[12],[13].
Figure 2: Determination of Normal Center of Gravity (shown with stars) Euclidean distance [11] between different facial points is calculated. In 2-D, the Euclidean distance between (x1,y1) and (x2,y2) is (x1-x2) ^2+ (y1-y2) ^2
(1)
Multilayer perceptrons with one input, one or more hidden layers and one output layer is the necessary condition in back propagation network. Networks that are being trained using backpropagation can have more than two hidden layers, which can make learning complex relationships easier for the network. Other architectures add more connections, which might help networks learn. The employed neural network is a feedforward multilayer neural network hidden layer as given in the following Figure 4.
This is the default method for calculating the Euclidean distance. We consider the Euclidean distance between different facial features. The distance between the eyes is calculated first then correspondingly the distance between all facial points is calculated.
Figure 4: Back Propagation Multi-layer neural network structure B. Error Back-Propagation Algorithm The weighting factor of the input-to-hidden neurons can be computed by (2)
wij( k 1) wijk Figure 3: Calculation of the Euclidean distance between different facial points.
E ( k ) wij
(2)
4
Where k is iteration number; i, j are index of input and
E hidden neuron, respectively; and η is step size can be wij
training face images and it is tested with rest of the testing images. In BP two hidden layers are used so BP is a Multilayer Neural Network (MLNN).
calculated from the following series of equations (3)-(6). The error function is given by
1 p E (t l ol ) 2 l 1
2
(3)
Where p is the number of output neurons, l is the index of neuron, tl and ol are the target and output values, respectively. The activation function, net function, and output function are given by (3)
si
1 1 e( neti )
(4)
Figure 5:Some of the facial images from grimace face database.
n
net i wil xl win1
(5)
l 1
m
oi vil sl vim1
(6)
l 1
Where n is the number of input neurons, and m is the number of output neurons. Let us define
E E si net i si net i
(7)
Comparison of the Morphological feature Method with standard algorithms i.e. Principal Component Analysis (PCA) [14],[15] and Regularized-Linear Discriminate Analysis(RLDA) [16],[17]. The experiment shows that it behaves slightly better than the PCA but it is slightly poor than (R-LDA) when trained with Back Propagation Algorithm. R-LDA and PCA methods are performed on Grimace Face database with 360 color image. The optimum configuration of BP neural network for PCA, R-LDA and our method used for training & testing is shown in Table-I. Table-I : BP neural network Configuration
And
E E net i wij net i wij
(8)
then we obtain the weight update equation (2) for the input-tohidden layer by computing Eq. (7) and Eq. (8) with the Eqs. from (3) to (6). Next, vij, hidden–to–output neurons’ weight update can also be derived in the same way.
IV.
EXPERIMENTAL RESULTS
A. Environments The experiment is performed on the Grimace Face database, which contains 360 coloured face images of 18 individuals. There are 20 images present for each subject. Database images vary in expression & position. But due to the robustness of images (images taken under different lighting and images taken on different days) we consider 240 colored image that can be used in the experiment. In this experiment, 120 images i.e. 8 from each selected folder are used for training data set & rest (120 images) are selected as testing set. The original images are preprocessed to extract the features. First the images are converted to gray scale. Then on the converted images segmentation techniques are used. B. Classification by the Back Propagation Network The Euclidean distances between different fiducial points are fed as input t the BP algorithm. BP is applied to the
BP-NN configuration
for PCA
for R-LDA
Morphological feature Extraction Method
Input Vector Nodes
20
14
6
Number of Hidden Layers
2
2
2
Number of neurons (hidden layer 1 ,hidden layer 2 & output layer) Transfer functions (hidden layer 1 , hidden layer 2 & output layer ) Network Learning rate
30,35,18
29,29,14
31,37,15
Tansigmoid tansigmoid, linear
Tansigmoid tansigmoid, linear
Tansigmoid, tansigmoid, purelin
0.001
0.001
0.0001
V. CONCLUSION At first glimpse the result of Morphological feature Method was somewhere between 60% and 70%. This was much lower than results reported so far. On Analysis carefully we explore that this was due to robustness of our selected images[18]. But after considering the best images for the experiment the result was of satisfactory level.
5
Training graphs of BP-NN applied to PCA, R-LDA and Morphological feature Method preprocessed training set are shown in Figure 6,7 and 8.
Careful observation from results denote that, The recognition performance using Morphological feature Method with BP-NN is superior to the performance of the PCA with BP-NN as per the Table II. Recognition performance using R-LDA with BP-NN is superior to the performance of the Morphological feature Method with BP-NN as per the Table II. 94 93.5 PCA + BP
93
RLDA + BP 92.5
MM + BP
92 recognition rate
Figure 6: Learning of BP-NN after preprocessing by PCA.
Fig 20: Recognition rate of PCA+BP, RLDA+BP and Morphological feature Method (MM) +BP-NN Limitation and Future work
Figure 7. Learning of BP-NN after preprocessing by R-LDA
Face detection itself is a challenge. We have presented in this paper a novel technique of face recognition that is independent of the aging factor, orientation, pose, facial expression and presence of accessories like glasses, beards, etc by considering the fiducial points. The But it has certain limitations of recognition under various lighting conditions. At first glimpse, the results on test data was in between 60% and 70%. This was much lower than results reported so far. On Analysis carefully we explore that this was due to robustness in lighting conditions in selected images. Further research are proposed to develop person identification numbers (PINs) on the basis of these fiducial points. PIN’s system is more accurate and inexpensive. As the PIN’s are individualized character so can’t be stolen by someone. There is no need of carrying or remembering them like passwords. The proposed system focused on still images, a video-based face recognition provides several advantages over still imagebased face recognition[19],[20].
Figure 8:Learning of BP-NN after preprocessing by Morphological feature Method PCA, R-LDA and Morphological feature Method preprocessed input vectors training results are shown in table II. Methods
PCA+ BP-NN 33.08 sec. 13
R-LDA+ BP-NN 41.5 sec
We are thankful to The Director of ABV-IIITM for support & facilities provided to complete this work. We would also like to express our appreciation to the producer of Grimace database http://cswww.essex.ac.uk/mv/allfaces/index.html
Morphological Method + BP-NN 82.46 sec
Overall execution time No. of error 11 9 images Recognition 92.77% 93.88% 93.33% Rate (167/180) (169/180) (112\120) Table II: Recognition Rate and Execution time using PCA,R-LDA and Morphological feature Method with BP
ACKNOWLEDGMENT
REFERENCES [1] [2]
[3]
Intelligent Face Recognition Techniques: A Comparative Study. GVIP Special Issue on Face Recognition, 2007 Q. Liu, X. Tang, H. Lu, and S. Ma, “Face recognition using kernelscatter-difference-based discriminant analysis,” IEEE Trans. Neural Netw., vol. 17, no. 4, pp. 1081–1085, Jul. 2006. W. Zheng, X. Zhou, C. Zou, and L. Zhao, “Facial expression recognitionusing kernel canonical correlation analysis (KCCA),” IEEE Trans.Neural Netw., vol. 17, no. 1, pp. 233–238, Jan. 2006.
6 X. Tan, S. Chen, Z. H. Zhou, and F. Zhang, “Recognizing partially occluded,expression variant faces from single training image per personwith SOM and soft k-NN ensemble,” IEEE Trans. Neural Netw., vol.16, no. 4, pp. 875–886, Jul. 2005. [5] De Vel, O.; Aeberhard, S., “Line-based face recognition under varying pose”.Pattern Analysis and Machine Intelligence, IEEE Transactions on , Volume: 21 Issue: 10 , Oct. 1999, Page(s): 1081 -1088. [6] Sumanth TammaDecember 5, 2002S. Face Recognition Techniques. [Online]. Available: http://ie.emu.edu.tr/admin/dosyalar/%7Be3F-az2Rta%7Dneura.pdf [7] B. L. Zhang, H. Zhang, and S. S. Ge, “Face recognition by applyingwavelet subband representation and kernel associative memory,” IEEE Trans. Neural Netw., vol. 15, no. 1, pp. 166–177, Jan. 2004. [8] Rama Chellappa, Charles L. Wilson and SaadSirohey, “Human and machine recognition of faces:A survey,” Proceedings of the IEEEvol.83,no.5.pp.705-740 May1995. [9] Mehmoud Ismail, Reda A. El-Khoribi “HMT of theRanklet Transform for Face Recognition andVerification,” ICGST’s Graphics Vision and Image Processing Journal, vol.6,issue3,dec06,pp7-13. [10] Debasis Mazumdar, Santanu Dutta and Soma Mitra “Automatic feature detection of a face and recovery of its pose” Communicated to [4]
Journal of IETE. [11] Breu, Heinz, Joseph Gil, David Kirkpatrick, and Michael Werman, "Linear Time Euclidean Distance Transform Algorithms," IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 17, No. 5, May 1995, pp. 529-533. [12] http://en.wikipedia.org/wiki/Backpropagation [13] http://page.mi.fu-berlin.de/rojas/neural/chapter/K7.pdf [14] K. V. Mardia, J. T. Kent, and J. M. Bibby, Multivariate Analysis. Academic Press, 1979. [15] M.A. Tanner, Tools for Statistical Inference: Methods for the Exploration of Posterior Distributions and Likelihood Functions. Springer, 1996. [16] Version 5.0 The CSU Face Identi_cation Evaluation System. Web address:http://www.cs.colostate.edu/evalfacerec/algorithms5.html. 2003. [17] R. O. Duda, R E. Hart, and D. G. Stork, Pattern Classification. WileyInterscience, 2001 [18] K. Delac, M. Grgic, S. Grgic, Statistics in Face Recognition: Analyzing Probability Distributions of PCA, ICA and LDA Performance Results, Proceedings of the 4th International Symposium on Image and Signal Processing and Analysis, ISPA 2005, Zagreb, Croatia, 15-17 September 2005, pp. 289-294 [19] W. Zhao, R. Chellappa, A. Rosenfeld and P.J. Phillips.Face Recognition: A literature survey. Technical Report CART-TR-948. University of Maryland, Aug. 2002 [20] [20] L. Torres, J. Vilà, “Automatic Face Recognition forVideo Indexing Applications”, Pattern Recognition, Vol 35/3, pp. 615-625, December 2001.
