Face Recognition System on Mobile using Similarity Data Mining Technique

This paper proposed the mobile face recognition using data mining technique system. This system is able to help the police to find the suspects from the crime database using Smartphones. The police has to mark the 10 important points in face that are 2 of beginning of the eye lids, 2 of end of the eye lids, nose tip, 2 of center of pupils, 2 of end of mouths, and chin prominence. This system is divided into 2 main functions which are creating the database, and recognizing the face in database. From the recognition process, this research used the data mining technique called “K-nearest neighbor technique”. Our experiment is tested on the 100 pictures students from Department of Computer Science, Faculty of Infomatics, Mahasarakham University. The experiments are divided into 2 tests which are normal user test and software developer test. The normal user test is tested by 5 users with 30 problems, and the average percentage of true detection is 60.664%. The software developer is tested on 3 times of 100 problems, and average percentage of true detection is 65.667 %.


Introduction
The face recognition is vital to many tasks such as Facebook face tag, face defendant for police, security for example access control buildings, ATM machines, Image database investigation, Smart card.The face recognition consists of 2 processes that are face detection and face recognition.Many research interested in this topics.Xiaoguang (1) proposed the image analysis for face recognition using model-based approaches including Elastic Bunch Graph matching.Rabia and Hamid (2) proposed the survey of face recognition techniques.The two primary tasks of face recognition is verification (one-to-one matching), and identification (one-to-many-matching).The usually features used for face recognition is ratios of distance, areas and angles.Shang-Hung (3) proposed an introduction to face recognition technology.The framework for face recognition system consisted of 4 main processes that are face detector, eye localizer, facial feature extractor, and face recognizer.Zhimin et al (4) proposed the face recognition with learning-based descriptor.The components are forehead, left eyebrow, right eyebrow, left eye, right eye, nose, left cheek, right cheek, and mouth.Tim et al (5) proposed the principles and methods for face recognition and face modelling using adding the active shape model, skin textures.Yoshihisa et al (6) proposed the security management for mobile devise by face recognition using Bayes law.The feature is variation of genders and ages.
Apart from the previous research, this research tries to use the simple face recognition for mobile because of low processor.The user will touch the 10 importance points on face image.After that, the distance between that points will automatically calculate and send to the server.The server will recognize with the database using Similarity Analysis.

Face Recognition System on Mobile using Similarity Data Mining Technique
comparing the similarity with the database.This system consisted of 4 steps that are Pre-processing, Feature Extraction, Model generating, and Evaluation as shown in Figure 1 and Figure 2.

Pre-Processing
This research interested to use in mobile.The normal resolution of image from mobile is 4 to 5 megapixel.Then, this research firstly collects the 100 peoples of 5 megapixel images in JPEG format.The image is in clearly white background.Width and Height face ratio will detect by face tacking.The object in the various focal lengths such as short distance, long distance has different size.However, it is very difficult to calculate the focus length in mobile device, then this research selects to use the width and height of face ratio to normalize the position of (x,y).The x ' and y ' can be calculated by equation 1 and 2. The width and height of face will show in Figure 3.

Feature Extraction
From Figure 3, the user must mark the 10 important points in face that are 2 of beginning of the eye lids, 2 of end of the eye lids, nose tip, 2 of center of pupils, 2 of end of mouths, and chin prominence as shown in Figure 4.The example of the feature will show in Table 1.
Table 1.The features for each image.

Model Generating and Model Usage
The simple recognition is try to compare the difference of the Euclidean distance between unknown and all record in database.This algorithm is similarity analysis called k-nearest neighbors algorithm (7)(8) .This algorithm is simple, easy-to-implement, fast, and sophisticated machine learning methods.The algorithm can be summarized as: 1.A positive k is specified where k is the number of nearest output.The performance is primarily determined by the choice of k.This research sets to 3.
2. The identification of face recognition is calculated by equation 4, then compare with the all records in database.
where Q n is the feature of unknown P n is the feature in the database n is the number of features.
3. Output is the k lowest distance record.

Evaluation
This paper will test the data using percentage of true detect normal calculated by equation 5.

Percentage of true detect normal
=    * 100 (5 where TP is the true output Total normal is the number of test.

Experimental Result
This research developed the algorithm on Android platform, and tested with the 2 experiments that are tested by user, and tested by developer.The main application will show in Figure 5(a-d).The Figure 5 For tested by user, the 30 images will mark and test by 5 users, the average percentage of true detected rate is 60.66.The output will show in Figure 6.For tested by developer, the 100 images will test for 3 times, the average percentage of true detected rate is 65.66.The output will show in Figure 7.

Fig. 1 .
Fig. 1.Face Recognition System on Mobile using Data Mining Technique framework.

Fig. 4 .
Fig. 4. The 10 important points in face.The 10 positions will calculate the distance between all points by Euclidean (equation 3) that the features in this research are 45 features.
(a) is the main mobile application, and Figure 5(b) is the image with marking.The Figure 5(c) is the output and Figure 5(d) is the detail of selected output.