Download Person Authentication Using Hough Transform in Ear Biometrics PDF

Person Authentication Using Hough Transform in Ear Biometrics
Name: Person Authentication Using Hough Transform in Ear Biometrics
Pages: 14
Year: 2014
Language: English
File Size: 420 KB
Downloads: 0
Page 2

II. RELATED WORK The potential of the human ear for personal identification was recognized and advocated as early as 1890 by the French criminologist Alphonse Bertillon. Bertillon made use of the description and some measurements of the ear as part of the Bertillonage system that was used to identify prisoners . In his studies regarding personal recognition using the outer ear in 1906, Richard Imhofer needed only four different characteristics to distinguish between 500 different ears . In 1949, the American police officer Alfred Iannarelli conducted the first large scale study on the discriminative potential of the outer ear. He collected more than 10,000 ear images and determined 12 characteristics needed to unambiguously identify a person. Figure 1 show the measurement used in this approach. Fig 1. a. Ear anatomy: (1) helix rim, (2) lobule, (3) anti helix, (4) concha, (5) tragus, (6) anti tragus, (7) crux of Helix, (8) triangular fosse and (9) incisor (10)intertragica b. 12 Measurements used The problem is to detect the presence of groups of collinear or almost collinear figure points in the ear image. It is clear that the problem can be solved to any desired degree of accuracy by testing the lines formed by all pairs of points. However, the computation required for n points is approximately proportional to n 2 , and may be prohibitive for large n. Also due to imperfections in either the image data or the edge detector, however, there may be missing points or pixels on the desired curves as well as spatial deviations between the ideal line and the noisy edge points as they are obtained from the edge detector. Hough proposed an interesting and computationally efficient procedure for detecting lines in images. Hough Transform exploited the point line duality to identify the supporting lines of sets of collinear pixels in images. The Hough transform (HT) is a used for line detection due to its robustness to noise and missing data. Duda and Hart explored the fact that any line on the x y plane can be described in terms of and . Hence vertical lines which do not have a slope intercept equation can also be represented. In this representation, is the normal distance and is the normal angle of a straight line. Applying the Hough Transform to a set of edge points (x i , y i ) results in an 2D function C( , ) that represents the number of edge points satisfying the linear equation = x cos + y sin . The local maxima of C ( , ) are to detect straight line segments passing through edge points. It also shows how the method can be used for more general curve fitting. A method based on reduced Hough transform was proposed by Banafshe Arbab Zavar and Mark S. Nixon .It proposes the use of the Hough Transform (HT), which can extract shapes with properties equivalent to template matching and is tolerant of noise and occlusion. It finds the elliptical shapes in 2D face profile images to locate the ear regions by using a Hough transform to gather votes for putative ellipse centers in an accumulator, which will go through a refinement process that will eliminate some of the erroneous votes; the location of the peak in this accumulator gives the coordinates of the best matching ellipse. It achieved 91% enrollment success. Liu performed ear segmentation using histogram based K means clustering and Hough transformation for ear detection without using training dataset .The clustering is directly applied on to the 2D images. Pflug et al presented a survey on the approaches in 2D and 3D ear biometrics, covering ear detection and ear recognition systems, discussed their characteristics and reported their performance. The Table I summarizes . Annapurani .K et al. / International Journal of Engineering and Technology (IJET)ISSN : 0975 4024Vol 6 No 5 Oct Nov 20142040


Page 3

TABLE I Summary of Automatic ear detection methods for 2D and 3D images Authors Detection Method No. of Images and Type Performance Prakash & Gupta Connectivity graph 1604 3D 99.38% Arbab Zavar & Nixon Hough Transform 942 2D 91% Ansari & Gupta Edge detection & curvature Estimation 700 3D 93.34% Liu & Liu Adaboost and skin color filtering 50 2D 96% Suryaprakash, Gupta. P et al .The proposed technique is based on connected component analysis of a graph constructed using the edge map of the side face image .Ear localization then builds the edge connectivity graph, and subsequently uses it in finding the connected components in the graph for ear localization purpose. The components having the maximum connectivity are found to form the ear boundary. It produces a maximum success rate of 95% on dataset of 1070 images. The Figure 2 shows the geometric ear recognition performed in . Fig 2. Geometric ear recognition (a) Original edge image, (b) Edge image after approximating edges with line segments Vyas et al presented a survey of the ear detection methods.14 methods for 2D images and 3 methods for 3D images were discussed with their characteristics and reported their performance. The results found in are tabulated in Table II. Annapurani .K et al. / International Journal of Engineering and Technology (IJET)ISSN : 0975 4024Vol 6 No 5 Oct Nov 20142041


Page 4

TABLE II Accuracy of various ear detection methods III. EAR AUTHENTICATION SYSTEM This paper primarily aims at developing an ear based biometric system which can be used for authenticating people. The increasing variety of biometrics applications in everyday identification and authorization problems urges the development of easily applicable methods. Therefore rather than use specialized 3D equipment, the focus is for generic deployment via planar images derived from digital cameras. The ear is largely a planar shape and 3D must penetrate the intricate inner ear, which restricts deployment potential. In the case of 2D recognition, detecting ears from cropped side face 2D images is a challenging problem due to the fact that ear images can vary in appearance under different viewing and illumination conditions .A suitable way of finding the features of ear images under various environmental conditions by the use of the Hough transform on ear images is proposed here which is tolerant to rotation and illumination variation. This ear authentication system comprises three stages namely Image Acquisition and Preprocessing, Localization and Feature extraction using Hough Transform Matching In the preprocessing stage, the ear image is normalized and standardized to a common format. Then the feature extraction algorithm is applied on the ear image. Originally the edges belonging to ear are curved .But the Hough Transform (HT) breaks every edge of the outer ear into a set of line segments called Hough Lines. This makes it more compact for further processing. The HT arrives at a distance score. The users are assigned a unique secret code each .The distance scores generated and the secret codes are combined to form a feature vector for matching. The block diagram of the proposed ear authentication system is shown in the Figure 3.The authentication proceeds in two phases: Enrollment phase and Verification phase S.No Technique Database Used No. of Images Accuracy 1. Deformable contour based N/A N/A N/A 2. Outer Helix Curve Based IITK 700 ~93% 3. Template Matching N/A 103 N/A 4. Genetic Local Search N/A 110 N/A 5. Morphological Operations Based Method WVU 376 90% 7. Shape of low level Features XM2VTS 252 99.6% 8. Skin Color and Template Based IITK 150 94% 9. Distance Transform And Template matching IITK 150 95.2% 10. Active Contour Based N/A N/A N/A 11. Adaboost UND 203 100% 12. Shape based UND 142 100% 13. Method based on Reduced Hough Transform XM2VTS 252 91% 14. Histogram based K means Clustering and Hough Transform CVL 180 90% Annapurani .K et al. / International Journal of Engineering and Technology (IJET)ISSN : 0975 4024Vol 6 No 5 Oct Nov 20142042


Page 5

Fig 3. Block Diagram of the ear authentication system A. Enrollment phase Enrolment (or registration) in an automatic biometric system is the process of detecting and isolating the area of interest .The enrolment data record comprises one or multiple biometric references and other non biometric data such as a unique password. In subsequent uses, biometric information is detected and compared with the information stored at the time of enrollment. It consists of the three steps: Image acquisition, pre processing and Localization using Hough transform. B. Verification Phase First, in verification (or authentication) mode the system performs a one to one comparison of a captured ear image with a specific template stored in a biometric database in order to verify the individual is the person they claim to be. Two steps are involved in the verification of a person: : feature extraction and matching. In the first step, the images are subject to feature extraction to produce a feature vector. The feature vectors of enrolled image and query image are then matched. If the distance is less than a threshold, the person is authenticated. IV. ENROLLMENT PROCESS A total of 104 ear images of 26 individuals are enrolled, with each individual having 4 images captured under varying environmental conditions. One of the 4 images is the training image. The system is tested against the other three. A. Image Acquisition The side face images are acquired using touch less imaging set up under indoor conditions with illumination changes and at varying angles of rotation. All the images are taken from the right side of the face with a distance of approximately 20 25 cm between the face and the camera. The images have been stored in JPEG format. B. Pre Processing In this approach the ear part is manually cropped from the side face image and the portions of the image which do not constitute the ear are colored black leaving only the ear. But due to the noise in the image noisy edges may be detected only partially. Low pass filtering, otherwise known as "smoothing", is employed to remove high spatial frequency noise from a digital image is performed using the Wiener filter. The median filter is later applied to restore blurred edges and to remove the strong vertical edges which are part of image borders. They are created when the region containing the ear is cropped from the side face image. The edges of ear image are represented in white and the remaining area is made black by image binarization. Edge detection is performed using the Sobel operator . Query Image Image Pre processing Localization using HoughTransform Extraction of feature vector Computation of distance score + Secret code retrieval Matching Result Distance score for enrolled image + Secret code Annapurani .K et al. / International Journal of Engineering and Technology (IJET)ISSN : 0975 4024Vol 6 No 5 Oct Nov 20142043


Page 6

Fig 4. Edge detected image It can be seen that from Fig 4 the Sobel operator works best for the image as the edges are preserved and are not approximated to form a smooth curve.The outliers are removed properly. The edge detector has detected both horizontal and vertical edges in the ear image. C. Localization and Feature Extraction Using Hough Transform This approach proposes the use of the Hough Transform (HT), which can extract shapes with properties equivalent to template matching and is tolerant of noise and occlusion, to find the elliptical shape of the ears in 2D head profile images. The feature extraction stage prepares an acquired sample for matching and authentication by separating the object of interest from the background. It is an important step for any automatic biometric authentication system. The algorithm finds the elliptical shapes in cropped 2D side face images by using a Hough transform to gather votes for the region. The regions having maximum votes are plotted as edges. Such curved edges are fitted to straight lines forming a connected edge map. The problem therefore is decomposed into several constrained sub problems i.e. Hough lines .The smaller edges which do not form the boundary are removed leaving only the location of the outer boundary of the ear. Thus the problem of detecting collinear points can be converted to the problem of finding concurrent curves . D. Ear Localization and Approximation of Edges Using Line Segments Ear localization first builds the edge map, and uses it in finding the collinear points in the map for ear localization purpose. Hough transform is performed on an edge detected ear image. Each white pixel on the edge detected binary image is represented as a coordinate pair in parameter space ( , ) .For each edge point (x i, y i ), the value of is calculated as between each such cell along the theta axis is 1 degree and 1 degree along axis. A 2D array called accumulator A ( , ) is used to store the value of and for all edge points in a cell. A voting procedure is then carried out in the parameter space on A. For each edge point the accumulator is incremented At the end of this procedure, , means there are P points in this edge. A graph is plotted for the values of the accumulator bin,


Page 7

Fig 5. Sinusoidal curves in parametric form The Figure 5 shows the plot for the values of A. These are plotted for the lines with rotation 90 to +90 degrees. The point of intersection of these sinu soidal curves represents point of intersection of the lines in coordinate space. But they are mostly parallel corresponding to non intersecting lines on the ear boundary. The white boxes in the figure 5 show the peaks of the curves, which are the maxima of the curves. The maxima is found by thresholding the accumulator .A threshold of 50% is used. The values above the threshold are considered as peaks. These peaks correspond to the strong lines in the image. V. VERIFICATION PROCESS The straight lines thus found are fitted to the curvature of the outer ear forming an edge map as shown in Figure 6. The edges which are linear (or almost linear) need only two points for their representation after line segment fitting. In this manner the curved edges are broken down into a set of straight lines called Hough Lines. The Hough Lines are plotted on the binary ear image. The lines are green in color. Their endpoints are depicted in red and yellow Let the set of Hough lines of an image be . A. Extraction of Distance Score The Euclidean distance between end points of each Hough Line segment in is calculated .Let the distances so obtained be . The distance score is defined as The line is highlighted in blue in Figure 6 .The above procedure was performed for all the test images in the gray scale format and the result set stored for further matching and comparison. B. Matching First, in matching mode the system performs a one to one comparison of captured ear image with a specific template stored in an ear image database in order to verify the individual is the person they claim to be. It is a simple algorithm which can be used in security systems. 1) Ear Image Database: It contains the ear images of different people taken in advance. Images of 26 subjects having a resolution of 204x272 are taken with each subject having 4 images of varying rotation, intensity or illumination are taken and stored. 2)Secret code: Both the images are placed in separate folders with their names as a unique secret code assigned to every individual. Two images of the same individual have the same code. One of these is used as sample input and the other as reference image. 80 60 40 20020406080 300 200 100 0 100 200 300Annapurani .K et al. / International Journal of Engineering and Technology (IJET)ISSN : 0975 4024Vol 6 No 5 Oct Nov 20142045


Page 8

Fig.6 Result of Hough lines in MATLAB C. Security & Two Part Matching Firstly the secret code is submitted at the time of comparison. Any user cannot access the system without the knowledge of the secret code. This prevents fake login attempts. Let the secret code for a particular subject be .Let the distance score for the ear image be .The distance score and the secret code form a feature vector where is the feature vector of the query image . Let the feature vector of the reference image be, where is the secret code and Now the reference image submitted by the user during the enrollment phase, having secret code is fetched. The distance D between the two feature vectors is calculated as the Manhattan distance between the two distance scores, and as The distance D is calculated as the distance between the max lines provided their secret codes and 1' are identical, i.e.; the distance is found between the enrolled image of a particular user and the test image of a user who claims to be a legitimate user. D. Authentication If D is less than a threshold then a match is found. A threshold of 70% is used. If the distance D between two images is less than 70%, then the images are said to be matched. The person is verified to be who he/she claims to be. The results of the matching procedure are stored for further analysis. VI. RESULTS AND DISCUSSION This technique is tested against the IIT Delhi Ear database, with 26 enrolled users. Each subject has 4 images taken under different angles and illumination. Total of 104 images are enrolled. One image per subject is used as the training image and the test dataset contains three images per subject. The system prompts for the secret code and then the ear image. Annapurani .K et al. / International Journal of Engineering and Technology (IJET)ISSN : 0975 4024Vol 6 No 5 Oct Nov 20142046


Tags: Download Person Authentication Using Hough Transform in Ear Biometrics PDF, Person Authentication Using Hough Transform in Ear Biometrics free pdf download, Person Authentication Using Hough Transform in Ear Biometrics Pdf online download, Person Authentication Using Hough Transform in Ear Biometrics 2014 download, Person Authentication Using Hough Transform in Ear Biometrics.pdf, Person Authentication Using Hough Transform in Ear Biometrics read online.
About | Contact | DMCA | Terms | Privacy | Mobile Specifications
Copyright 2021 FilePdf