Face Recognition For Remote Database Backup System Aniza Mohamed Din 1 , Faudziah Ahmad 1 , Mohamad Farhan Mohamad Mohsin 1 , Ku Ruhana Ku-Mahamud 1 , Mustafa Mufawak Theab 2 1 Graduate Department of Computer Science,UUM College of Arts and Sciences, Universiti Utara Malaysia, 06010 UUM Sintok, Kedah, Malaysia 2 UUM College of Arts and Sciences, Universiti Utara Malaysia, 06010 UUM Sintok, Kedah, Malaysia Abstract- Face recognition is one of the most interesting applications in the image processing field. To build a model to recognize the face of different people, we need to do several processes on the image to obtain the most efficient features. In this research a face recognition model is developed. The dataset used is of different face images. Neural Networks technique, specifically Multilayer Perceptron (MLP) model with Back- Propagation learning algorithm and Template Matching approach are implemented in model developed. The face recognition model developed is then applied on a remote database backup system. Template matching approach is found to give a higher percentage of matching accuracy and a faster result can be obtained compared to MLP as no learning process is required I. INTRODUCTION Organizations need to have a good data backup strategy to prevent data loss. Unfortunately, saving files in databases does not guarantee safety from threats or disasters. Files in a database can be deleted by failure or accident, and data can be destroyed due to hard disk error or virus infection. Unexpectedly, computer can be physically destroyed from natural disaster such as fire or flood or even be stolen by maliciously act. Since data loss can be a very serious problem to an organization, data backup is an important routine. That is, to make one or more copies of the database files regularly and put them in a safe place, such as another machine or server. Organizations, such as banks deal with various transactions every day, where this information is critical. Business organizations must make backups on a consistent basis to ensure the safety of the transactions data. This means that it is essential for organizations to make backups at specific times regardless of location be it in-house or remotely. Backing up databases in the organization itself is less threatening than backing up databases remotely. In remote backups, greater security measures are needed. One method to enhance security measures for remote backup systems is incorporating facial recognition technology. A remote database backup system is where users can backup and compress their database servers remotely. If the application is run on a machine connected to LAN or WAN, all the servers‟ names will appear in the server list. Otherwise, the user can add a server name or IP manually. After a connection to a server is made, all the databases‟ names will be listed in the database list view and users can choose the database that they wish to backup. This paper proposes a remote database backup system using facial recognition technology. The aim of the system is to address current needs for reliable identification and verification of individuals. II. PREVIOUS WORKS Face recognition is a very interesting and difficult problem because the variations in the image brightness, different faces and different people expressions. Sometimes it is hard for the people themselves to recognize the difference between the people‟s faces, so to develop a program that can reach such objectives is very challenging. Many studies have been done in this area and several algorithms have been used and one of them is Neural Networks. Neural Networks (NN) can be implemented in many different applications to fulfill the user/s requirements and it has been a very popular tool in image recognition and data classification. The facial recognition developed employs Neural Networks method specifically Multilayer Perceptron (MLP) with Backpropagation (BP) algorithm. BP shows very strong ability to solve many complex problems in different domain. In order to apply Neural Networks on images (face images) an extraction methods should be applied first to extract the features from the images. In [9], there are two types of techniques to present input data in face recognition systems, i.e., the feature-based technique and the image-based technique. In feature-based technique, the 288 input data is merely a number of features extracted from the image, while in image-based technique the input data is the processed image itself. Usually, features extracted from grayscale images or gray-scale images themselves are used as the input data in face recognition systems. A study in [2] combined two basic face detection methods i.e. skin colorbased method and feature-based method. In this approach, both methods are used where the color features and face features are extracted. A study about face recognition using artificial neural networks is proposed in [5]. The approach consists of two phases which are the enrollment and recognition/verification. Images were captured using a webcam and stored in a local dataset. After that, the images were processed to extract the features using methods such as Histogram and Homomorphic. For classification, Multi-layer feed forward with Backpropagation algorithm was used. An accuracy of around 98% was able to be achieved which indicated that the study has produced a good model in term of classification accuracy. Another study using artificial neural networks was proposed in [3]. In the study, a new approach to model face images using a state space feature was presented. Feature extraction was performed from the grayscale images of the human faces. For classification activites, Multi-layer feed forward with Backpropagation algorithm was used. For training set, 200 images were used and testing was performed on the set. The model managed to obtain accuracy around 98%. The important point in the study is that dimensionality reduction was used on the data set which is useful to reduce processing time. A survey about the algorithms and techniques used in face recognition was provided in [4]. The study investigated many features extraction algorithm such as Edges, Texture, Skin color and shape. It also investigated many classification algorithm such as Eignface, Distribution-based, Neural Networks, Support Victor Machine (SVM), Naïve-Bayes classifier, Hidden Markov Model, and Information-Theoritical Approach. From the investigation, it is concluded that the following points could affect the classification accuracy: lighting conditions, orientation, pose, partial occlusion, facial expression, presence of glasses, facial hair, and a variety of hair styles. Another approach used for face verification is template matching as studied in [6]. This approach is performed using an edginess-based representation of the face image. Experiments were conducted using a set of face images with different poses (position of the face towards the camera) and different background lightings. The approach used is proved to be a promising alternative to other methods when dealing with problems with different poses and background lighting. A study by [8] used 30 standard face images, focusing on the eye regions as templates for face detection. Template matching approach is applied together with 2DPCA algorithm, an algorithm developed in [7]. The results of the experiment conducted produces accurate rate of face detection in a short time. III. METHOD There are three (3) phases in this research. Phase 1: Reference Database Construction Authorized individuals images are captured using a webcam and stored in .bmp format, 122 x 160 pixel, and 32 bit depth. The administrator/user can specify the number of images to be taken (20 is the default value) and the sensitivity value, that is used to control similarity acceptance. The images captured are in four different poses (position of face towards the camera), background and lighting conditions. Each set of individual images is stored in separate folders. The output for this phase is a database of authorized personnel images in four different positions. This database is known as the reference database. Phase 2: Development of Facial Recognition System The aim of this phase is to detect a face and verify users. Two methods, Neural Networks and Template Matching are used to produce models. Models that produce the highest percentage of accuracy will be chosen for development. For Neural, Networks, the image must first be transformed into gray scale image. When the image has been cut, its features can be extracted. The features extracted are Color Mean, Color Standard Deviation, Gray Mean, Gray Standard Deviation, Luminosity Mean, Luminosity Standard Deviation, Brightness Mean, Brightness Standard Deviation, Saturation Mean, Saturation Standard Deviation, Gabor Mean, Gabor Standard Deviation, Contrast Mean, Edge Detection, Energy, Entropy, Homogeneity Mean, Homogeneity Standard Deviation, Sobel Mean, and Sobel Standard Deviation. After the feature extraction process, a normalization technique can be applied on the data. Phase 3: Evaluation Evaluations can be conducted in three ways; scenario, operational and technological [1]. Scenario evaluation is to evaluate the overall capabilities of the entire system for a specific application scenario, designed to model a real-world environment and population. Operational evaluation is to evaluate a system in actual operational conditions. Technological evaluation is to determine the underlying technical capabilities of the facial recognition system. For this research, evaluation on the technological aspect will be conducted. Specifically the system will be evaluated for performance on accuracy. Other evaluations methods are not in the scope of this research. IV. PROPOSED SYSTEM The system architecture and the phases of development are shown and described here. 289 A. System Architecture The architecture of the remote database backup system is shown in Fig. 1. Computers connected with a webcam must be used to enable the face recognition system to function properly. The machine must also be connected to the LAN, WAN, or internet to enable the system to access the database servers remotely. The system administrator is the only person responsible to register the images of the users to enable them to access and use the system. User 3 User 1 User 2 Face Recognition application Database server1 Database server 2 Database server3 Wireless connection Webcam Fig. 1. Architecture of Remote Database Backup System B. Face Recognition System The system has been developed in two steps: detect user‟s image and verification. Detect user‟s image: A webcam that is attached to a computer will capture the user‟s image and stored temporarily. Verify user: In this step, the system will trigger every time a user wishes to perform database backup. Image of user‟s current position will be captured during login via a webcam. The image captured (test image) must be in the same format and size as the reference image. Two algorithms are used to verify the images. The template matching algorithm will match this image with the images in the reference database. This approach is an exhaustive matching process, which performs complete scan of source image and comparing each pixel with corresponding pixel of template. Therefore here, it will match the pixels between the test image and the reference image. If a match is found, the user can start performing backup on the desired database remotely. For neural networks algorithm, the features of the user‟s image will be extracted and normalized. This means that the image must be standardized in terms of size, pose, illumination, etc., relative to the images in the gallery or reference database. The diagrams for facial recognition steps using neural networks and template matching are shown in Fig. 2 and 3 respectively. Fig. 2. Facial recognition steps using neural networks algorithm Fig. 3. Facial recognition steps using template matching algorithm C. Database Backup System Modelling Database backup system is where users can backup and compress their database servers remotely as shown in the flow diagram in Fig. 4. If the application is run on a machine connected to LAN or WAN, all the servers‟ names will appear in the server list. Otherwise, the user can add a server name or IP manually. After a connection to a server is made, all the databases‟ names will be listed in the database list view and users can choose the database that they wish to backup. This application will generate the backup file in compressed format by default. For automatic backup, a user can set all the parameters similar to a manual backup. Then, check the check-box titled „Daily Auto Backup‟, where a time setting component will be enabled to set the time for daily backup. A report of the System Detect user‟s image Verify user Normalize data Extract features Detect user‟s image Verify user 290 scheduled backup dates and list or errors, if any, that occurs during the connection to server, database selection or backup failure can be generated for reference. Fig. 4. Database Backup System Flow Diagram In this paper, we are going to focus on the development of the face recognition system, which is the security aspect of the database backup system. V. FINDINGS Experiments were conducted to test the performance of both methods used. Images of all authorized personnel for the database backup server must be taken for the experiments. For each user, 20 images were captured via a webcam. For Neural Networks, after the feature extraction process is performed, the data must be prepared for the learning process where it will be normalized to the range from 0 to 1. For the learning process, Multilayer Perceptron with Backpropagation learning algorithm is employed where the number of input units used is 20 units, while the hidden units used is 10 units. Learning rate and momentum values applied is 0.1. A structure of a Multilayer Perceptron is shown in Fig. 5. Fig. 5. A Structure of a Multilayer Perceptron Model The data is trained for 5000 epochs or until the error rate is 0.001. The final weights of the model from the learning process must then be stored in a database. To test the performance of the model built, new images of the authorized personnel are captured via a webcam. The features of each image are extracted and normalized. The final weights stored are then used to classify the images. For this model, the percentage of accuracy for classification achieved is in the range of 70% to 75%. The low percentage of accuracy may be due to the variety of poses and background lighting captured in the images used in the training and testing phases. For template matching, no features extraction or learning process needs to be done. Images of all authorized personnel are captured via a webcam and stored in a database. Even though the number of images used in the experiments is 20 by default, in this method, the administrator can determine the number of images to be captured for each user. However, for x 1 x 2 X 20 Z 1 Z 10 Y 1 . . . . . . Input Layer Hidden Layer Output Layer 291 comparison purposes, the default value is used. It is important to note that the more images used, the more processing time taken during the process of image matching. The images of each personnel are stored in a separate location in the database. Those set of images are considered as templates or reference images. For testing, new images of the personnel are taken via a webcam. The image format must be in the same format as the templates, which are in .bmp format, 122 x 160 pixel, and 32 bit depth. However, the background, light and illumination can be different than those in the template images because a user could login from a different location and environment. Based on the sensitivity value specified to control the similarity acceptance during the matching process, the percentage of accuracy for the image classification is in the range of 80% to 85%. If a closer image of the face is captured, better accuracy can be achieved. The difference in performance is probably due to the ability of template matching to match any image with template images by doing a complete scan of a new image and comparing each pixel with the corresponding pixel of a template. Therefore, this technique is practical for a situation when we do not want to bother with features extraction and understand which features to be selected for certain type of images. Results can also be obtained in a short time as no learning process is required in this approach. VI. CONCLUSION This paper proposes a remote backup system using facial recognition technology. The aim of the system is to address current needs for reliable identification and verification of individuals. The facial recognition model is conducted in three phases: Reference Database Construction, Development of Facial Recognition System and Evaluation. Two algorithms Neural Networks and Template Matching are used to produce models. Models with the highest percentage of accuracy will be chosen for developing the remote database backup system. Template matching approach is found to give a higher percentage of matching accuracy compared to MLP. Results can also be obtained in a short time as no learning process is required. This research will extend the literature on face recognition domain. ACKNOWLEDGMENT This research is supported by the Leadership Development Schemes (LEADS) grant. We thank Universiti Utara Malaysia and the Ministry of Higher Education Malaysia for financing the research. REFERENCES [1] L. D. Introna, and H. Nissenbaum, “Face recognition technology: a survey of policy and implementation issues”, Center for Catastrophe Preparedness and Response, New York University, 2009. [2] N. Jamil, S. Lqbal, and N. Iqbal, “Face recognition using neural networks”, Proceedings of IEEE INMIC 2001, IEEE International Multi Topic Conference 2001: Technology for the 21st Century, pp. 277 – 281, 2001. [3] V. Kabeer, and N. K. Narayanan, “Face recognition using state space parameters and artificial neural network classifier”, Proceedings of International Conference on Computational Intelligence and Multimedia Applications, pp. 250-254, 2007. [4] Y. Ming-Hsuan, D. J. Kriegman, and N. Ahuja, “Detecting faces in images: a survey”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(1), pp. 34-58, 2002. [5] S. A. Nazeer, N. Omar, and M. Khalid, “Face recognition system using artificial neural networks approach”, Proceedings of International Conference on Signal Processing, Communications and Networking (CSCN '07), pp. 420-425, 2007. [6] A K. Sao and B. Yegnanarayana, “Face verification using template matching”, IEEE Transactions on Information Forensics and Security, vol. 2, no. 3, pp. 636-641, September 2007. [7] J. Wang and H. Yang, “Face detection based on template matching and 2DPCA algorithm”, IEEE Congress on Image and Signal Processing, pp. 575-579, 2008. [8] J. Yang, D. Zhang, and J. Yang, “Two-dimensional PCA: A new approach to appearance-based face representation and recognition”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 26(1), pp. 131-137, 2004. [9] K. Youssef, and W. Peng-Yung, “A new method for face recognition based on color information and a neural network”, Proceedings of The Third International Conference on Natural Computation (ICNC 2007), 2007. 292