WWW.DISSERTATION.XLIBX.INFO
FREE ELECTRONIC LIBRARY - Dissertations, online materials
 
<< HOME
CONTACTS



Pages:     | 1 |   ...   | 2 | 3 || 5 | 6 |   ...   | 33 |

«Ear Recognition Biometric Identification using 2- and 3-Dimensional Images of Human Ears Anika Pflug Thesis submitted to Gjøvik University College ...»

-- [ Page 4 ] --

• Throughput and accuracy: The search result should be delivered within a reasonable amount of time (several minutes). However, accuracy is still more important than throughput. A lower false accept rate (the probability that the wrong subject gets a high rank in the candidate list) is preferred over a lower false reject rate (the probability that the true perpetrator does not show up in the candidate list), because a false accept would imply that an innocent subject could be accused for a crime.

2. THE GES-3D PROJECT Figure 2.1: Floor plan for the data collection in GES-3D. Reference data is collection with a special setup that consists of 3 depth cameras (left). Probe data is collected by four CCTV cameras.

2.1 General Requirements GES-3D is a project where partners from research and industry collaborate with the goal of creating a demonstrator system. In order to protect the interests of the industry partners, it is important to make sure that their software can be integrated into the system as a black box. Concurrently running modules should run independently from each other and at the same time, the interface should leave as much freedom as possible for the individual solutions of each project partner (w.r.t. algorithms, libraries and programming languages).

In GES-3D, the exchange of data is implemented with a proprietary internal interface.

Each project partner delivers a back-end module that provides the functionalities that are specified in this interface (see Figure 2.9). We distinguish between the data capture subsystem [89] and the system back-end. The capture system is operated by a police officer and used for enrolment. The retrieval system is operated by a specialized department of the criminal police. Both parts of the system connect to a central web service, by implementing the previously mentioned interface. Figure 2.2 illustrates the two parts of the system. For Enrolment, we obtain a series of mugshots and a 3D model (left part) and for identification, we obtain one or several videos (right part). The search function also accepted single video frames or images.

The prospective end user of the GES-3D system is the German Federal Criminal Police Office (BKA). BKA is operating the prospect identity retrieval system and will also deploy the image capture system to the local police stations. The process of enrolment is defined by national and international standards and follows a strict protocol [134, 33]. It is crucial that the system fully reflects these standards and seamlessly integrates into the existing work flow. The 3D capture system also collects a series of mugshots (see next section for details) from different poses in addition to the 3D head model. This allows a smooth migration from the existing identity retrieval system (GED-2D) to the new system.

2.2 IMAGE CAPTURE SYSTEM

–  –  –

Figure 2.2: Division of tasks with associated media types in the proposed forensic identification system.

2.2 Image Capture System Interpol and national police agencies in Europe and the US [134] have adopted this standard. The standard is designed to enable the exchange of images between national and international police stations. It is also needed for establishing a quality standard of the images in the forensic databases. According to the standard, it is recommended that at least one frontal image and two profile images with 45 and 90 degrees of pose variation are captured. The profile images should show the left and the right ear. Further, it is recommended that the focal length of a camera should be between 85 mm and 135 mm and that the mid-points of the mouth and of the bridge of the nose should be centered. Shadows should be minimized by using at least two sources of diffuse light. We follow these recommendations and use a data set with five images which represent the previously described poses. The images are captured with a high resolution digital camera under the described environment for each subject in the enrolment and compressed with JPEG compression.

In order to maximize the probability of identifying a subject, the quality of the input media for enrolment should be as high as possible. In particular, there should be no compression artefacts in the 2D images. The 3D models should be as detailed as possible, while containing a low noise level and a minimum of reflection artefacts. 3D face imaging is well proven and tested in various scenarios. 3D ear imaging, however is a new application, which poses some challenges to the arrangement of the capture devices. On one hand, it should be possible to obtain high quality face images and on the other hand, the angles between capture devices should be arranged in a way that we obtain a complete view of the concave and self-occluding surface of the ear.

For obtaining a high quality representation of the outer helix, the capture protocol was designed accordingly. We capture two partial ear models, one with a full profile view and another one that shows the ear from behind at approximately ±120 degrees (see the two left sub figures in Figure 2.4 for an illustration of these two camera angles). This allows us to obtain a smooth representation of the outer helix. We also get information on the perturbation of the ear.

2. THE GES-3D PROJECT Figure 2.3: Example images for a collection of mugshots, as they are collected during the enrolment process.





Figure 2.4: Different views of an example for an interpolated 3D head model, as it is collected during the enrolment process with a close-up view of the ear.

2.3 Dataset The collection of a dedicated dataset was an objective of the GES-3D project. The data was collected in accordance with the standards and recommendations described in the previous section. The GES-3D corpus contains 300 distinct subjects, from which 150 are males and 150 are females. 113 subjects were wearing glasses and 98 subjects wore earrings. The age distribution ranges from 20 until more than 60 year old subjects.

All subjects were asked to enter a room, where an ATM was placed next to a wall. The dotted line in 2.1 represents the expected path of the subjects from the door to the ATM.

The room is also equipped with four CCTV cameras, which are marked in green and with numbers 1-4 in Figure 2.1. One camera was placed next to the door, the second one was placed on top of the ATM, a third cameras was placed at the opposite side of the room with respect to the door and a fourth camera was placed next to the ATM’s screen. Each video camera is running at a resolution of 1920x1080 pixels and captures 30 frames per second.

Example images for each of the camera viewpoints are shown in Figures 2.5, 2.6, 2.7 and 2.8.

The 3D images were captured with a setup of three viSense2 depth cameras (upper left corner in Figure 2.1). The camera uses structured light for capturing the surface of an object at a scanning speed of 0.25 ms. The diameter of the frustum for the 3D acans is 782x582 pixels and the resolution of 2D camera for capturing the texture data is 1624x1234 pixels.

Each depth camera is connected with several flashes in order to assure optimal illumination settings. Drop shadows are minimized by using flash diffusers. The partial depth images are registered semi-automatically by an operator. After capturing all input images the operator is asked to annotate three corresponding landmarks in each partial model. These landmarks can be chosen arbitrarily by the user as long as she is able to coarsely point to the same landmarks in two corresponding 3D models. It is recommended to use landmarks that can easily be found in every model, such as the nose tip, the outer corner of the eye

2.4 SYSTEM BACK-END or outer corner of the lip. The registration algorithm performs a coarse alignment based on these manually marked points followed by an automatic fine alignment of the surfaces using Iterative Closest Point Search (ICP) [27]. The output of this is a full 3D model of the subject’s head (see Figure 2.4 for an example).

The 2D mugshots are collected simultaneously with the partial 3D models. For each subject, we obtain a left and right full profile view, a left and right half profile view and a frontal view. The mugshots are taken with a high resolution digital camera using JPEG compression (1624x1234). The enrolment data collection setup is separated by a thin wall with a back cloth, in order to obtain mugshots with a uniform background. An example for the mugshots is shown in Figure 2.3.

The reference images are stored together with some meta data, such as a unique identier for the subject, a unique identifier for the capture sessions, and a time stamp for the 2D and for the 3D images. In our evaluations, we use data from 200 randomly chosen subjects.

The remaining 100 subjects are reserved for black box testing.

The collection, storage and processing of the data is subject to the German data and privacy protection regulations, which state that the data may not be published without the explicit consent of each subject and the data may not be used for any other purpose but the GES-3D project. In order to maintain reproducibility and compatibility of our results, we decided to use public datasets rather than the GES-3D data in our scientific publications.

2.4 System Back-End

From a technical perspective, the system back-end consists of five different subsystems, which are the biometric database, the back-end provider module, several biometric service providers (encapsulating the algorithm-specific feature extraction and retrieval methods), the fusion service provider, a connection for the biometric capture device and a module for handling search requests (see Figure 2.9). Together, these subsystems follow the work flow specification for a general biometric system as specified in ISO/IEC SC37 SD11 [89].

The central component is the back-end provider where all intercommunication between the modules is processed. The middleware JBoss Application Server is used to connect the various interfaces between the different modules from the project partners.

For enrolment, the input data is read from the capture device and passed over to each biometric service provider. Triggered by this event, each provider initiates the enrolment process and returns a feature vector. The back-end provider then stores each feature vector in the biometric database, where it is available for future search operations. The processing of the data in our module is written in MATLAB, OpenCV and Java EE with additional external libraries.

The overall system is supposed to mimic the central server that is run by the criminal police. It takes images from the capture device, sends them to the biometric service providers for feature extraction, stores templates and performs search operations upon request. For data retrieval, video sequences are sent to the service provider. Depending on the pose and possible occlusions, the face and ear region are segmented and a template is generated from these regions. Because the database contains a complete 3D model of the head for each subject, pose variations in the input images can be compensated during the comparison process.

Each biometric provider generates a proprietary template. The algorithms for enrolment are developed independently from each other and the extracted feature vectors reect different properties of the image. A separate fusion component combines the results from each provider and generates a ranked list of identities. The list is sorted according to a descending likelihood that the subject in the probe and the reference sample have the same identity.

2. THE GES-3D PROJECT Figure 2.5: Example images for camera viewpoint 1. The camera was placed next to the entrance door.

Figure 2.6: Example images for camera viewpoint 2.

The camera was placed on top of the ATM Figure 2.7: Example images for camera viewpoint 3. The camera was placed at the opposite side of the room with respect to the door.

Figure 2.8: Example images camera viewpoint 4.

The camera was placed next to the screen of the ATM.

–  –  –

Figure 2.9: Overview of the architecture of the GES-3D identification system with connections to the database back end, biometric service providers and the capture device.

2.5 Workflow of a Biometric Service Provider A biometric service provider is connected with the back-end provider via a WSDL-API. It combines functionality from OpenCV and MATLAB code with Java and can be deployed on a JBoss 7.1 compatible web service. The WSDL file defines method stubs for the enrolment and the search method, as well as the exchange formats for input media and search results. The interface also provides functionality for retrieving data from the database via the back-end and specifies a unified format for templates and search results. The data retrieval and database access is handled by the back-end provider module. All biometric service providers automatically obtain the same probe data and have fully transparent access to the central database via the back end provider module. Additional libraries are specified in a Maven file and linked to the JBoss environment.

A general overview of the workflow of our provider showing the combination between OpenCV, MATLAB and Java is illustrated in Figure 2.10. For an illustration of overall the segmentation, preprocessing, feature extraction and feature vector generation process, please also refer for Figure 2.13.

2.5.1 Enrolment The enrolment task takes at least one input medium and returns one feature vector per input medium to the back-end module. Possible media types for enrolment are a series standardized photographies, as described in Section 2.2 or a 3D head model.

2.5.2 Segmentation The method for ear segmentation depends on the type of the input media. Possible media are a series of photographs, a video stream or a 3D head model.

Series of mugshots: The preprocessing module segments the ear region from the half and full profile images using the Viola-Jones detection method [182] with the implementation from OpenCV (this implementation uses the set of wavelets suggested in [182]). We trained our own haar cascade for detection using manually cropped positive images from the

2. THE GES-3D PROJECT

–  –  –



Pages:     | 1 |   ...   | 2 | 3 || 5 | 6 |   ...   | 33 |


Similar works:

«HORIZON 5 (1) 2016 : I. Research : J. Vydrová : 214–231 ФЕНОМЕНОЛОГИЧЕСКИЕ ИССЛЕДОВАНИЯ • STUDIES IN PHENOMENOLOGY • STUDIEN ZUR PHÄNOMENOLOGIE • ÉTUDES PHÉNOMÉNOLOGIQUES 1.4 PHENOMENOLOGY AND AESTHETICS 1.4 ФЕНОМЕНОЛОГИЯ И ЭСТЕТИКА doi: 10.18199/2226-5260-2016-5-1-214-231 THE INTERTWINING OF PHENOMENOLOGY AND CUBISM — IN THE ANALYSES AND WORKS OF ART OF CZECH ARTISTS AND THEORETICIANS JAROSLAVA VYDROVÁ PhD, Scientific fellow...»

«Torah Praxis after 70 C.E.: Reading Matthew and Luke-Acts as Jewish Texts by Isaac Wilk Oliver A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy (Near Eastern Studies) in The University of Michigan 2012 Doctoral Committee: Professor Gabriele Boccaccini, Chair Professor Raymond H. Van Dam Assistant Professor Ellen Muehlberger Assistant Professor Rachel Neis Professor Daniel Boyarin, University of California, Berkeley To my Father, Benoni...»

«PAssAgewAy imPliCit •  into the PASSAGEWAY INTO THE ImPLICIT Nada Lou Focusing as a philosophical practice has gone through a fine, long nourishing period of being sustained inside the psychotherapy milieu. Its therapeutic values were recognized, developed, acknowledged, used, written about and esteemed. Then, at the end of the millennium “Thinking at the Edge” (TAE) appeared at the edge of the familiar Focusing scene. At first many of us felt bewildered: “What is Gene doing to my...»

«THE EFFECTS OF THE EQUATION OF STATE AND THE COMPOSITION ON THE THERMAL EVOLUTION OF NEUTRON STARS by Mehmet Tuncer Unver A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Physics MONTANA STATE UNIVERSITY Bozeman, Montana May 2013 ©COPYRIGHT by Mehmet Tuncer Unver All Rights Reserved ii APPROVAL of a dissertation submitted by Mehmet Tuncer Unver This dissertation has been read by each member of the dissertation committee and has been...»

«ABSTRACT Title of Dissertation: BECOMING A RECOGNIZED HISPANICSERVING INSTITUTION (HSI): INSTITUTIONAL CULTURE & TITLE V Rebecca C. Villarreal, Doctor of Philosophy, 2014 Directed By: Associate Professor Noah D. Drezner, Department of Counseling, Higher Education, and Special Education The purpose of this study was to understand the influence of institutional culture at an HSI on the Title V program. The findings highlight the forces that influence the process of securing, evaluating, and...»

«The effects of habitat and predation on bay scallop populations in New York A Dissertation Presented by John Michael Carroll to The Graduate School in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Marine and Atmospheric Science Stony Brook University August 2012 Stony Brook University The Graduate School John Michael Carroll We, the dissertation committee for the above candidate for the Doctor of Philosophy degree, hereby recommend acceptance of this...»

«Piedras votivas de Pampacolca Nuevos datos sobre las lajas pintadas del sur del Perú Votivsteine aus Pampacolca Neue Daten über bemalte Steinplatten aus Südperu Inauguraldissertation zur Erlangung des Grades eines Doktors der Philosophie am Fachbereich Geschichtsund Kulturwissenschaften der Freien Universität Berlin vorgelegt von Renata Faron-Bartels M.A. Berlin 2011 1. Gutachter: Prof. Dr. Jürgen Golte 2. Gutachterin: Prof. Dr. Ursula Thiemer-Sachse Tag der Disputation: 10. 11. 2009 2...»

«The Experience and Self-Management of Fatigue in Adult Hemodialysis Patients by Ann Elizabeth Horigan Nursing Duke University Date:_Approved: _ Julie Barroso, Supervisor _ Diane Holditch-Davis _ Susan Schneider _ Sharron Docherty Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Nursing in the Graduate School of Duke University ABSTRACT The Experience and Self-Management of Fatigue in Adult Hemodialysis Patients by Ann Elizabeth Horigan...»

«Male Education and Son Preference in India by Rebha Sabharwal A Dissertation Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy Approved July 2013 by the Graduate Supervisory Committee: Sarah Hayford, Chair Victor Agadjanian Scott Yabiku ARIZONA STATE UNIVERSITY August 2013 ABSTRACT The study of son preference in India has been the focus of research for a few decades. The desire for sons leads to unfavorable consequences for daughters such as unequal access...»

«Szab-Ch09.qxd 09/07/04 21:17 Page 356 Naming and Asserting Scott Soames Many essays in semantics and the philosophy of language seem to proceed on the assumption that—special circumstances involving ironic, metaphorical, or other non-literal uses of language aside—the proposition asserted by an utterance of a sentence in a context is the proposition semantically expressed by the sentence in that context. At some level, of course, we all know that this is a fiction, since sometimes a single...»

«THE CUSTOMER’S PATH TO LOYALTY: A PARTIAL TEST OF THE RELATIONSHIPS OF PRIOR EXPERIENCE, JUSTICE, AND CUSTOMER SATISFACTION by Denver Eugene Severt Dissertation submitted to the Faculty of the Virginia Polytechnic Institute and State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY in Hospitality and Tourism Management APPROVED: Dr. Suzanne Murrmann Dr. Cherylynn Becker Dr. Roberta Minish Dr. Pamela Weaver Dr. John Williams April 23, 2002...»

«Physics College of Science Graduate Programs 2014-1015 Contents Faculty 1 The Graduate Programs 5 The Doctor of Philosophy Degree 6 The Master of Science Degree 15 Our Mission: To educate students for a life of fulfillment and accomplishment. To Create and translate knowledge to meet global and societal needs. Northeastern University is an equal opportunity/affirmative action Title IX education institution and employer. Tuition rates, all fees, rules and regulations, courses, and course content...»





 
<<  HOME   |    CONTACTS
2016 www.dissertation.xlibx.info - Dissertations, online materials

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.