Advanced Ear Biometrics for Human Recognition

Transcription

1 International Journal of Computer Systems (ISSN: ), Volume 03 Issue 11, November, 2016 Available at Sandhiya B Asst.Professor, IT Department, Easwari Engineering College, SRM Groups, Chennai , India Abstract Human ear has attracted huge researchers attention because of its uniqueness, and stability. Clearly ear edge detection and template creation and matching plays a very big role in human ear recognition method. In this paper, canny edge detector method is used for edge detection and template matching algorithm technique with sobal filter is used for template matching to improve ear recognition rate and its accuracy is proposed. The proposed system provides a detection rate of 99.9% and an identification rate of 95.4% with better accuracy has obtained. Results have been compared with existing techniques. The results demonstrate the effectiveness of technique for human identification. Keywords: Biometrics, Canny edge detector, Template matching technique, Edge detection, Ear Recognition. I. INTRODUCTION Biometric is the science of identifying or verifying the identity of an individual and it has the capability to reliably distinguish between an authorized person and a phony. Nowadays, there are numerous technique (password, pins etc) to classify and verify the identity of a person. Biometric, offers much higher authenticity than the traditional methods. An ideal biometric must be universal, unique, permanent and collectable. Ear biometric has advantages over the other recognition technologies since its structure is not changing during the lifetime of an adult and is unaffected by facial expression unlike other biometrics (face, eyes etc). Human identification has been a subject of intensive research for the few decades because of its applications in almost all aspects of secure surveillance. Biometric systems have become very essential components in almost all security aspects. Biometrics deals with recognition of individuals based on their physiological or behavioral characteristics. Researchers have done extensive studies on biometrics such as fingerprint, face, palm print, iris, and gait. Ear, a viable new class of biometrics, has certain advantages over face and fingerprint, which are the two most common biometrics in both academic research and industrial applications. For example, the ear is rich in features; it is a stable structure that does not change much with age and it does not change its shape with facial expressions. For identical twins also some Eigen vector changes will be there. So it s more effective vision of biometrics. Furthermore, ear is larger in size compared to fingerprints but smaller as compared to face and it can be easily captured from a distance without a fully cooperative subject although it can sometimes be hidden with hair, cap, turban, muffler, scarf, and earrings. The anatomical structure of the human ear is shown in Figure 1. Figure-1 Anatomy of Human Ear The ear is made up of standard features like the face. These include the outer rim (helix) and ridges (antihelixes) parallel to the helix, the lobe, the Concha (hollow part of ear), and the tragus. II. BACKGROUNDS AND RELATED WORKS Ear was first used as a human biometric by Iannarelli [4] who compared more than 10,000 ears samples for study and found that, structure of ear does not change radically over time and having the ability to be consider as human biometric like figure print or iris. The medical literature [4] provides information that ear growth is proportional after first four months of birth and changes are not noticeable in the age 8 to 70. We have studied some other author s research work based on two dimensional ear images and summarize them as follows. Yuan and Mu [11], in this paper, used a normalization method based on the concept of an improved Active Shape Model (ASM). Ear normalization was adjusted for any scaling and rotational variation in image. Then Full-space Linear Discriminant Analysis (FSLDA) was applied to perform ear recognition and achieved a recognition rate of 90%. According to Xie and Mu, multi-pose problem erupts 619 International Journal of Computer Systems, ISSN-( ), Vol. 03, Issue 11, November, 2016

2 only when the angle between the subject ear and the camera changes, causing the distortion. Chang, Bowyer, Sarkar and Victor, built a recognition system by taking the help of both face and ear. The technique used by them was PCA. They manually pass two coordinates of the triangular fossa and the antitragus. There on PCA was used to extracting features point known as earspace [10]. Burge and Burger [9], transformed the subject ear into the model of adjacency graph. The graph construct was based on Voronoi diagram which is further derived from the use of Canny extraction based on curve segments. They designed a graph matching logic for authenticating a person. Zhang and Liu, analyzes the problem of multi view ear ecognition. They used B-spline pose manifold construction in a discriminative projection space. This space is formed by the Null Kernel Discriminant Analysis (NKDA) feature extraction scheme. They reported a 97.7% rank-1 recognition rate [13]. Sana and Gupta [20], they extracted the structural features of the ear by using Haar wavelet transforms. The Haar wavelet transform was applied to separate the discovered subject image and to calculate coefficient matrices of the wavelet transforms which are clustered in its feature template. The correctness of their algorithm was 96%. Nosrati et al. [22], they applied a 2D wavelet on an aligned ear image. Template matching algorithm was used for feature extraction. The features was diverged in various positions (horizontal, vertical, and diagonal). They merged these lost images to create a single feature matrix. They achieved a recognition correctness of 90.5%. III Image Database I EAR DATABASE SOURCE Purpose: Supporting academic research of ear recognition Subject: Students and teachers from the department of Information Engineering, USTB. The total number of volunteers is 60. Condition: The right ear is photographed with digital camera Detail: Every volunteer is photographed three different images. They are normal frontal image, frontal image with trivial angle rotation and image under different lighting condition. Each of them has 256 gray scales. Images had already experienced rotation and shearing, but they were without illumination compensation Image Database II Purpose: Supporting the academic research of ear recognition methods, particularly under lighting and angle variations, as well as ear image pre-processing methods Subject: Students and teachers from the department of Information Engineering, USTB. The total number of volunteers is Condition: The subject s head in right hand view is photographed by CCD camera. The distance between subject and camera is fixed to 2 meters Detail: Every volunteer is photographed four images. They are profile image, two images with angle variation and one with illumination variation. Each image is 24-bit true color image and 300*400 pixels. The first image and the fourth one are both profile image but under different lighting. The second and the third one have the same illumination condition with the first while they have separately rotated +30 degree and -30 degree with the first one. Thus, the main purpose of the image database is to support the research about ear recognition under illumination variations and angle variations Image Database III Purpose: Supporting the research concerning about steps of ear recognition system including ear detection, the robustness of recognition methods under depth variation, ear recognition under partial occlusion and multi-modal biometric feature recognition based on the fusion of information from ear and face Subject: Students and teachers from the department of Information Engineering, USTB. The total number of volunteers is IIT Delhi Ear Database The IIT Delhi database of an ear image is gathered by taking the snaps of students and staff ear present in the campus. The acquisition of an image was done during Oct 2006 Jun 2007 using simple imaging set up. The age groups of the subjects are in between years. The resolution of an image is 272 x 204 pixels in jpeg format. 3.5 IIT Kanpur Ear Database It consist of two set of data i.e. Dataset -1 and Dataset- 2. Data Set 1 has 801 side face images acquired from 190 subjects. Data Set 2 has again 801 side face images collected from 89 subjects. It consist of frontal view of the ears captured at three positions, first when a person is looking straight, second when person is looking approximately 20down and third when person is looking approximately20 up. IV. PROPOSED WORK In my work, I have used USTB data base-i, a set of 60 subjects. Average of three ear images of each subject was taken, as 180 ear images of 60 subjects was used. 4.1 Capture Ear Image First step of the work is to collect standard data set. Therefore, either create own data set or used standard dataset, which are available on various universities. Here digital camera or web camera is used as a hardware device to capture ear image. You can get the data set on special request from the university authority. In order to create own data set, of capturing ear images, various factors, we have to be keep in mind like, Stability of camera and its height. Distance between camera and subject. Subject should be stable. Constant light ambience. 620 International Journal of Computer Systems, ISSN-( ), Vol. 03, Issue 11, November, 2016

3 Complete feature extraction method summarize in following steps. Change format of ear image, from RGB to Grayscale. Set all the parameters of canny edge Detector function on specific values. Pass grayscale image as a parameter in canny edge detector and get unique ear signature in return and save it as a binary image. Figure 3: Captured Image 4.5. Feature Extraction Process using Canny Edge Detector Complete feature extraction method described in Figure Proposed System Process Flow Complete method is summarized in following steps; RBG Image Grayscale Image 4.3 Preprocessing Normalization Figure 4: Proposed System Steps Once, data set collected, it should be normalized before processing. Here normalization means that all the images must be equal in dimension. If the images are not equal, then they have to alter in a standard dimension through cropping and resizing. As we have used USTB data set-i for the processing. All the images are equal in dimension. 4.4 Feature Extractions The ear is made up of standard features like fingerprints. In biological context, ear includes the outer rim (helix) and ridges (antihelixes) parallel to the helix, the lobe, the concha (hollow part of ear), and the tragus (the small prominence of cartilage over the meatus). All these biological features of ear, we have used as a unique signature in the form of edges. In order to extract ear features in the form of edges, we have used most popular method of edge detection named as Canny Edge Detector. Ear Signature Figure 5: Feature Extraction Process Extraction of feature vector There are two feature vectors which are under consideration. Both these are taken from the outer edge of the ear to reduce the computational complexity and to minimize the errors from the feature extraction process. Here to find a reference point the concept of max line is used. In this a line which is the maximum possible distance from any two pixels of the outer edge is find out. Now the midpoint is found out which is the reference point. 621 International Journal of Computer Systems, ISSN-( ), Vol. 03, Issue 11, November, 2016

4 For the 1st vector division of the line into 20 points is done.normal from these points are taken and their intersection with the boundary is stored. Angles from the reference point and the point obtained are stored as 1st feature vector. An optimal value of n for the division of line should be consider for satisfying the Accuracy, space and time requirement. For 2nd feature vector reference point is taken as the midpoint of the normal which is taken from the main reference point. A line is drawn from the 2nd reference point to the top of max line 1 and is considered as the 2nd reference line. Now for the 2nd feature vector same steps as taken for the 1st feature vector are taken. No. of Eign Vectors Figure 6: Feature Vector Points 4.7 Ear Feature Vector Recognition A database of 1st and 2nd feature vector of the subjects who are considered for comparison is as follows: 1st feature vector is compared with the entire subjects 1 st vector; if it is greater than a threshold then further comparison is made. All subjects whose 1st feature vector match is compared for 2nd vector matching and all where match is greater than threshhold2 are shortlisted. Now subject who has maximum vector point matches is displayed as matching subject. b. Find the gradient strength and direction with: G G 2 x G G arctan( G 2 y y x ) The direction is rounded to one of four possible angles (namely 0, 45, 90 or 135) 3. Non-maximum suppression is applied. This removes pixels that are not considered to be part of an edge. Hence, only thin lines (candidate edges) will remain. 4. Hysteresis: The final step. Canny does use two thresholds (upper and lower): a. If a pixel gradient is higher than the upper threshold, the pixel is accepted as an edge b. If a pixel gradient value is below the lower threshold, then it is rejected. c. If the pixel gradient is between the two thresholds, then it will be accepted only if it is connected to a pixel that is above the upperthreshold. Canny recommended a upper:lower ratio between 2:1 and 3:1. VI. CLASSIFICATION Classification is the task of finding the match for a given query image. V. CREATING AN EDGE BASED TEMPLATE MODEL USING TEMPLATE MATCHING Create a data set or template model from the edges of the template image that will be used for finding the pose of that object in the search image. Use variations of Canny s edge detection method to find the edges Proposed Algorithm For edge extraction, Canny uses the following steps: 1. Filter out any noise. The Gaussian filter is used for this purpose 2. Find the intensity gradient of the image. For this, we follow a procedure analogous to Sobel: a. Apply a pair of convolution masks (in X and Y directions). Figure 6: Classification Process. VII. PROPOSED METHODOLOGY The overall process includes image capturing in that phase image acquisition will be done, in preprocessing normalization takes place. After that feature extraction by canny edge detector and template matching techniques was done. 622 International Journal of Computer Systems, ISSN-( ), Vol. 03, Issue 11, November, 2016

5 Figure 9: Average Recognition Rate over No. of Training Iterations. Figure 7: Overall System Process The proposed method is implemented in MATLAB 7.5 on a PC with 2.27 GHz Intel processor and 3 GB RAM. Finally sobal filter was applied and classification takes place from the templates lastly ear recognition will be done. VIII. EXPERIMENTAL RESULTS In figure 8, average recognition rate from the number of Eigen vectors was obtained. And In figure 9, average recognition rate over number of training iteration was obtained. IX. CONCLUSION AND FUTURE WORK In this paper, a new method of human recognition is proposed based on template matching algorithm technique. Ear images are cropped and needs to be resized followed by conversion into grayscale image. After that Canny edge detector is used to extract the feature from the image. Database images are trained and stored in the form of average ear image as a template. Results obtained are promising and encouraging with 100% correct recognition rate. Although the ear biometrics is rich in characteristics but there are still some problems that need to be worked on to make automatic ear recognition system more effective and efficient in real world applications. In future for speeding up the search process further, a pyramidal approach can be used. Our future work would be extending the algorithm for rotation and scaling. This can be done by creating template models for rotation and scaling and performing search using all these template models. Figure 8: Average Recognition Rate over No. of Eigen Vectors In this experiment, ear recognition Success rate is 100% with 60 templates when applied ear query image out of 180 images of data set-in 10 number of training iteration. If the number of ear images increases in template formation then recognition rate would be improve. REFERENCES [1] Basit, A., Javed, M. Y. And Anjum, M. A., Efficient iris recognition method for human identification, ENFORMATIKA, pp , vol 1, [2] Moreno, B., Sanchez, A., Velez, J., F., On the Use of Outer Ear Images for Personal Identification in Security Applications, IEEE 33rd Annual International Carnahan Conference on Security Technology, pp , [3] Jain, A., Hong, L., Pankati, S., Biometric Identification, Communications of the ACM, vol. 43, No. 2, pp , [4] Iannarelli, A., in: Ear Identification, Paramont Publishing, 1989 [5] Victor, B., Bowyer, K., and Sarkar, S., An Evaluation of Face and Ear Biometrics, Proc. 16th Int l Conf. Pattern Recognition, pp , [6] Durgesh Singh*, Sanjay K. Singh A Survey on Human Ear Recognition System Based on 2D and 3D Ear Images OPEN JOURNAL OF INFORMATION SECURITY AND APPLICATIONS ISSN (Print): ISSN(Online): , Volume 1, Number 2, September International Journal of Computer Systems, ISSN-( ), Vol. 03, Issue 11, November, 2016

6 [7] Chen, H., and Bhanu, B., Contour Matching for 3D Ear Recognition, Proc. Seventh IEEE Workshop Application of Computer Vision, pp , [8] Raposo R, Hoyle E, Peixinho A, Proenca H. 'UBEAR: Adataset of ear images captured on-the-move in uncontrolled conditions'. In: Computational Intelligence in Biometrics and Identity Management (CIBIM), 2011 IEEE Workshop on; 2011.p [9] Carsten Steger, Markus Ulrich, Christian Wiedemann, Machine Vision Algorithms and Applications. [10] Digital Image Processing [Rafael C. Gonzalez, Richard Eugene Woods] el.edu/_weg22/can_tut.html. [11] USTB, University of science and technology beijing USTB database. Available at:http : www1:ustb:edu:cn=resb=en=index:htm. [12] UND, University of notre dame UND databases. Available at:http : ==www3:nd:edu= cvrl=cv RL=Data Sets:html. 624 International Journal of Computer Systems, ISSN-( ), Vol. 03, Issue 11, November, 2016