Face Tracking System with Haar Method and Pre-Study Face Recognition with Histogram Comparison

Size: px

Start display at page:

Download "Face Tracking System with Haar Method and Pre-Study Face Recognition with Histogram Comparison"

Brett Ramsey
6 years ago
Views:

1 Face Tracking System with Haar Method and Pre-Study Face Recognition with Histogram Comparison Endah Sudarmilah 1, Adhi Susanto 2 1 Department of Informatics, Muhammadiyah University of Surakarta Jl. A Yani Tromol Pos 1 pabelan Kartasura Telp. (0271) Fax (0271) Surakarta 2 Electrical Engineering, Engineering Faculty, Gadjah Mada University 1 endah_te@yahoo.com Abstract Face detection and recognition research rises with assumption identity, feature and character information of someone have extracted from images. Although commercial application for face detection and recognition have implemented, this technology not perfect yet, it need to develop for best result. Based on the background, problem can be pointed how to design and make automatic identification system with new method of face tracking and recognition. Haar algorithm has used in the research to face tracking and histogram comparison for face recognition. Microsoft Visual C compiler, C++ language, and image processing library OpenCV from Intel have used in this research too. Face tracking have specialized for frontal face not the others. The result gives 80 percent success for face tracking and recognition without changing light and structural component. Lightening factor and face expression haven t influence face tracking, since camera can capture face. Structural component haven t influence either, since tracking process hasn t lose the feature and vice versa. The size of tracking image depends on the distance of the object and camera, so much longer so much the smaller. Recognition process depends on lightening because it use histogram algorithm which get image intensity value. System can be used for every lightening condition if it has stagnant lightening for capture and database image. Face recognition can be used for every background according skin face histogram from tracking image. Index Terms haar algorithm, face tracking, face recognition, histogram comparison. I. INTRODUCTION Face recognition technology and the others biometrics technology (iris scanning, retina scanning, sound recognition, finger print, hand and fingers geometrics, signature verifications, walk style, ears or the others part of body). Whereas it is still strange, but it will develop dramatically as mention in MIT Technology Review [1], biometrics will be top ten emerging technologies that will change the world. This method developing was supported by computer technology, especially computer processor and video, makes implementation of face recognition will be large, and it is not impossible will growth as urgent application on every system that needs it [2], Face computing model is active research area since 1980 because this area not only in theory domain but also practice application in face recognition, such as crime identification, safety system, image and film processing, human-computer interaction and the others thing. Whereas computing model developing for face recognition is difficult, because of complex, multi dimension, and dynamics of face. Research about face recognition raise with assumption that information about identity, status and character of everyone can be extracted from images. The other assumption is computer that has reaction as their image vision. Topics of face recognition and expression interested researchers. w, whereas commercial application of face recognition have implemented [3], but this technology is not perfect yet [1] so this research still need to be developed, one of the research is face tracking system with haar method and pre-study face recognition with histogram comparison. II. FACE TRACKING Face tracking is detection and tracking face features with camera and marked as tracked image as output, in this step system recognize pattern as face or not. The process will be did by compiler Microsoft Visual C and image processing library OpenCV from Intel and Haar method with statistics approximation [4]. Training of Haar statistics model use positive sample which have face features and negative sample without it, both of these are trained together and the difference is used as face classification parameter. This information is saved and compressed as statistics model parameter which mentioned as "a cascade of boosted tree classifier". This process was known as haar training algorithm that had result file xml as statistics model parameter. This research used haar training parameter haarcascade_frontalface_alt.xml that was especially for

front face tracking. Process of face tracking with classifier file xml was shown at Fig. 1 and Interface of face tracking software can be seen at Fig. 2.

It compared with grayscale histogram from images in data base used three method comparison such as correlation, Chisquare, and intersection method (Open Source Computer Vision Library Reference

Tracked as a face? File xml (classifier) Face marked Tracked image Fig. 1. Flowchart of face tracking algorithm with classifier file xml Fig. 7.

Interface of face tracking software face skin image histogram III.

to the criteria, image which was shown is the original pixel value. On the other hand, value was not included to the criteria, image was shown is black (pixel value = 0).

Face recognition is face identification process based on face image that was saved in data base, this step system give output whose is this face?

2 front face tracking. Process of face tracking with classifier file xml was shown at Fig. 1 and Interface of face tracking software can be seen at Fig. 2. Image from Webcam Haar Tracking histogram comparison. This process found face image histogram first, that grayscale histogram. It compared with grayscale histogram from images in data base used three method comparison such as correlation, Chisquare, and intersection method (Open Source Computer Vision Library Reference Manual, 2001), all of this value used as conditional of face recognition with were ranged before. Flowchart of face recognition with histogram comparison algorithm can be seen at Fig. 4. Tracked as a face? File xml (classifier) Face marked Tracked image Fig. 1. Flowchart of face tracking algorithm with classifier file xml Fig. 7. Image histogram data and its comparison result Before images which were compared its histogram, tracked image take sample area of face skin image Fig. 2. Interface of face tracking software face skin image histogram III. FACE RECOGNITION Pre processing was did to limited tracked image with value of back projection of face skin image histogram (sample image was taken by selection of mouse), If value was included to the criteria, image which was shown is the original pixel value. On the other hand, value was not included to the criteria, image was shown is black (pixel value = 0). This pre processing can be shown at Fig. 3 and interface of face tracking application can be seen at Fig. 5 to Fig. 8. Face recognition is face identification process based on face image that was saved in data base, this step system give output whose is this face? Or face not recognize (face not exist in data base). Method which is used is find value of back projection Captured tracked image image Value limited of Histogram back project tracked image Histogram Comparison Value limited of back project Include include to the range? criteria? Recognize Pixel = 1 Stop original pixel value Unrecognize Image Data base Histogram Pixel = 0 Fig. 4. Flowchart of process of face recognition with histogram comparison Captured image Fig. 5. Sampled skin color histogram and Back projection image Fig. 6. Captured image as result of face tracking and its Fig. Fig. 3. Flowchart of face image pre processing algorithm histogram 8. Face recognition result

3 TABLE I ACHIEVEMENT OF RESULTS OF FACE TRACKING EXPERIMENT Criteria Condition Number of Samples Achievement Error 1 Face Position Lighting Good Dark Image Size Structural Component Face Expression False Positive False Negative would be pre processed with eliminated the background of tracked face images. IV. EXPERIMENT RESULT A. Experiment of Face Tracking Process Experiment of face tracking was did by face tracking process with different condition of face position, lighting, image size that was influenced with face distance from camera, structural component factor, and face expression, so will be known achievement of experiment results and how far the face tracking process can be done. In this The results of lighting factor to the face tracking process show that this criteria was not significant to influence the process with value of achievement was 1 or without error from 10 samples, but its quality still be influenced with camera catch. This achievement was supported by training factor that ignored color and image limited [5]. The criteria of tracked image size or object (face) distance from camera can be analyzed that this tracking have minimum and maximum distance, the minimum distance is the shortest length interval object to camera which camera can not track the object and the maximum is the most far length interval of it. The minimum distance is not be significant because of dependent to camera specification, for this research use experiment was taken 10 to 15 samples in each condition, for false positive and false negative was done by all of samples of experiment was about 70 samples. The achievement of this process can be seen at Table. I. Based on experiment results with criteria of frontal face position and non frontal face position, such as left diagonal, right diagonal, up face, down face and the other position the achievement result was 0.7. It indicate that this process is specified to face image with frontal face position to the camera. It can be proved that file xml was set up to 0.6 for x axis rotation, 0 for y axis rotation, and 0.3 for z axis rotation [5]. centimeters and maximum distance is about 150 centimeters from the camera. It also can be proved that this system was trained with minimum image size was 20x20 pixel and maximum size was 200x200 pixel [5]. Structural components which used in this experiment were glasses, face make up, moustache, beard, and veil or the other object that blocked the face. The structural components which were not lost recognize feature was not influence tracking process in this experiment for instance glasses. It can happen for the other structural components. Face expression didn t face tracking in this system, it can be proved in the experiment with various expression. Expression experiment got 10 samples and 100% can be tracked, although in discrepancy of expression. It was TABLE II ACHIEVEMENT OF RESULTS OF FACE RECOGNITION EXPERIMENT Criteria Condition Number of Samples Error Achievement 1 Face image file exist in data base exist t exist Good Lighting Medium Dark Image size Larger Smaller Structural components Face expression Logitech 5000 that the minimum distance is about 15 caused by training for the system that was used a lot of

4 face image [5]. Meanwhile, this experiment must fulfill the rule frontal face position. False positive is error that can happen in face tracking while the system detected a non face object as a face, and false positive is vice versa, while the system can not detected a face object as a face. It occurred commonly in every method of face tracking included haar method in this research. False positive was caused by there was the similarity of haar features in tracked image and face image was trained, and false negative was caused by technical factors of face tracking such as face position, structural components and size of tracked images. tracking in condition the exist of face images in data base, lighting, discrepancy of tracked image size (object distance to camera), structural components, and the difference of expression, can be its results in Table. II. Firstly, achievement value in criteria of the exist of images in data base was caused by the difference of histogram itself. Histogram is shown the spread of intensity of image pixel [6]. The difference of histogram will have result histogram comparison value which have different value too, and face recognition was depended on these. B. Experiment of Face Recognition Process Face recognition experiment process with face The second criteria was the level of lighting, in this experiment the lighting was not measured with a device but with estimated condition of lighting, for good condition was sampled room condition in afternoon time (lighting is good), medium condition was in night time but with good lighting from lamp, and the worse condition was in night time and with bad lighting from the lamp. Result of face recognition which was tested by lighting of face image, the more similar lighting condition in reference image and tracked image was, the higher its achievement value was in experiment. The difference of image size also was tested with two condition, which were tracked image was smaller than reference image, and tracked image was larger than reference image. Based on data of the result can be concluded that the difference of size of tracked and reference image had influence to the face recognition while it was not too significant. It was caused by face tracking process itself which must adapt with location of object. If the distance of object to camera was near, the image size that was saved had smaller size. On the other hand, the distance of object to camera was far, the image size that was saved had larger size. It could happen because the face tracking image was set with minimum size of tracked image was 20x20 pixel and maximum size was 200x200 pixel [5], so the system will adapt to the size of the object. Structural component was influenced to face tracked image, but if it still could be tolerated by haar algorithm, face will be tracked. Structural component including glasses, face make up, moustache, and beard, but structural component that could be tolerated by haar algorithm, face will not be tracked which is the veil that blocked haar features on face. Recognition system can not run if tracking system didn t happen. Face expression had influence to the face recognition but not really significant because face expression didn t make in histogram change. In this experiment used four condition of face expression namely, happy, sad, angry and surprise, and face images samples which were taken 10 times every expression were given result that high achievement. It was supported by [7] that histogram can developed by features for increasing quality to object recognition. C. Benefits and Drawbacks of System This face tracking and recognition with its experiment result that was did in this research have some benefits and drawbacks. These can used for adaptation to system application. The benefits there are: a. Face tracking with haar algorithm has good achievement 80% as a result. b. Face recognition with histogram comparison have simple algorithm and computation caused the process can run more quickly. c. System can used in every background condition because the face recognition use tracked face image with background elimination. d. Face recognition was adapted to the color skin face histogram. e. System can used in every lighting condition but both of lighting condition in tracked face image and data base image mat be same. The drawback of system are: a. Because of histogram comparison algorithm, face recognition depend on image intensity. b. System available only in the specific room with stagnant lighting condition. c. Color skin face image selection depend on an operator. V. CONCLUSION Based on design, build, and test face tracking and recognition system, could be conclude that are face tracking process specific for frontal face position, the

5 same lighting condition in tacked image and data base image must needed because system uses histogram comparison algorithm, structural components on face, image size and face expression are not influenced the recognition since it do not lost haar features and system can used in every background condition because the face recognition use tracked face image with background elimination. REFERENCES [1] J. Woodward, Horn, Gatune, &Thomas, Biometrics: A Look at Facial Recognition, Virginia State Crime Commision, [2] M. Yang, Kriegman, &Ahuja, Detecting Faces in Images: A Survey, IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 24,no. 1, 2002, pp [3] F. Fraser, Exploring The Use of Face Recognition Technology for Border Control Applications, Biometric Consorsium Conference, [4] P. Viola, &Michael J, Rapid Object Detection using a Boosted Cascade of Simple Features, IEEE CVPR, [5] N. Seo, Tutorial: OpenCV haartraining (Rapid Object Detection With A Cascade of Boosted Classifiers Based on Haar-like Features), PukiWiki Plus, [6] R. Munir, Pengolahan Citra Digital dengan Pendekatan Algoritmik, Informatika Bandung, [7] O. Linde, &Lindeberg, Object recognition using composed receptive field histograms of higher dimensionality, ICPR Proceedings of the 17th International Conference, Volume 2, August 2004, Page(s):1-6 Vol.2.

Progress Report of Final Year Project

Progress Report of Final Year Project Project Title: Design and implement a face-tracking engine for video William O Grady 08339937 Electronic and Computer Engineering, College of Engineering and Informatics,