`
`Alcatel-Lucent @
`
`BELL LABS
`_
`,! Technical Journal
`
`* a
`
`aot
`
`General Papers
`
`Volume 15, Number 1
`June 2010
`
`KAD
`KS
`io
`/
`\
`\
`
`\uaas
`
`(WY)WILEY
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 001
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 001
`
`

`

`
`
`EDITORIAL STAFF
`Editor-in-Chief
`Jane M. DeHaven
`
`
`
`Managing Editor
`Eileen M. Kammerer
`Editorial Associate
`Rebecca Nadolny
`Executive Editor
`Isabelle Cohen-DeAngelis
`John Wiley & Sons, Inc.
`Journal Production
`Jocelyn Kwiatkowski
`Wiley Periodicals, Inc.
`EDITORIAL BOARD
`ROD C. ALFERNESS, Chairman
`Chief Scientist
`Alcatel-Lucent Bell Labs
`Murray Hill, New Jersey USA
`SHARAD JAISWAL
`Technical Manager
`Alcatel-Lucent Bell Labs Applications Research Domain
`Bangalore, India
`KRISTIN F. KOCAN
`Technical Manager, Next Generation Systems and
`Software Applications Department
`Alcatel-Lucent Applications Software Group
`Naperville, Illinois USA
`STEVEN K. KOROTKY
`Distinguished Memberof Technical Staff
`Optical Networking Research
`Alcatel-Lucent Bell Labs
`Holmdel, New Jersey USA
`KURT LOESCH
`Memberof Technical Staff
`Alcatel-Lucent Bell Labs Germany
`Stuttgart, Germany
`OLIVIER MARTINOT
`Research Manager, Hypermedia Applications
`Alcatel-Lucent Bell Labs
`Villarceaux, France
`SAPE J. MULLENDER
`Director, Network Systems Laboratory
`Alcatel-Lucent Bell Labs
`Antwerp, Belgium
`THOMASPFEIFFER
`Department Head, Optical Access Research
`Alcatel-Lucent Bell Labs
`Stuttgart, Germany
`DAVE C, ROBINSON
`Strategy Director, Network Technology & Strategy
`Alcatel-Lucent Bell Labs
`Antwerp, Belgium
`JOSEPH A. TARALLO
`Director of CDMA andLTE BaseStation Systems, Engineering,
`Architecture, RF Performance and Technology Department
`Alcatel-Lucent Wireless Networks Group
`Murray Hill, New Jersey USA
`EVE VARMA
`Director of Standardization, CTO
`Alcatel-Lucent Optics Product Division
`MurrayHill, New Jersey USA
`THOMAS H. WOOD
`Technical Manager
`LGSInnovations, Government Communications Laboratory
`FlorhamPark, New Jersey USA
`
`
`
`June 2010
`
`The Bell Labs Technical Journal(Print ISSN: 1089-7089; Online ISSN:
`1538-7305 at Wiley InterScience, www.interscience. wiley.com)
`is published quarterly, four issues per year, by Wiley Subscription Services,
`Inc., A Wiley Company, 111 River Street, Hoboken, NJ 07030-5774.
`Copyright © 2010 by Alcatel-Lucent. Published by Wiley Periodicals, Inc.,
`A Wiley Company.All rights reserved. Nopart of this publication may be
`reproduced in anyform or by any means, except as permitted under
`section 107 or 108 of the 1976 United States Copyright Act, without either
`the prior written permissionof the publisher or authorization through the
`Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923;
`Tel: 978-750-8400; Fax 978-646-8600. The copyright notice appearing at
`the bottomof the first page of an article in this journal indicates the
`copyright holder’s consent that copies may be madefor the personal or
`internal use of specific clients, on the condition that the copier pay for
`copying beyond that permitted by law. This consent does not extend to
`otherkinds of copying, such as copying for general distribution, for
`advertising or promotional purposes, for creating collected works, or for
`resale. Such permission requests and other permission inquiries should be
`addressed to the Permissions Dept., c/o John Wiley & Sons, Inc., 111 River
`Street, Hoboken, NJ 07030-5774; Tel: 201-748-6011; Fax: 201-748-6008;
`http://www.wiley.com/go/permissions.
`Subscription price (Volume15, 2010): Print only: $295.00 in U.S.;
`$335.00 in Canada and Mexico; and $369.00 outside North America.
`Electronic only: $295.00 worldwide. A combination price of $325.00 in
`U.S.; $365.00 in Canada and Mexico; and $399.00 outside North America.
`Personal rate: $135.00 in the U.S., Canada, and Mexico; $159.00 outside
`North America. All subscriptions containing a print element, shipped
`outside the U.S., will be sent by air. Payments must be madein U.S. dollars
`drawn on a U.S. bank. Claims for undelivered copies will be accepted only
`after the following issue has been received. Please enclose a copy of the
`mailing label. Missing copies will be supplied whenlosses have been
`sustained in transit and where reserve stock permits. Please allow four
`weeksfor processing a change of address. Address subscription inquiries to
`Subscription Manager, Jossey-Bass, a Wiley Imprint, 989 MarketStreet,
`San Francisco, CA 94103-1741; Tel: 888-378-2537, 415-433-1767
`(International); E-mail: jbsubs@jbp.com.
`Postmaster: Send address changesto Bell Labs Technical Journal,
`Subscription Distribution U.S., c/o John Wiley & Sons, Inc., 111 River
`Street, Hoboken, NJ 07030-5774.
`Advertising Sales: Inquiries concerning advertising should be forwarded
`to Advertising Sales, c/o John Wiley & Sons, Inc., 111 River Street,
`Hoboken, NJ 07030-5774; Tel: 201-748-8832. Advertising Sales, European
`Contact: Jackie Sibley, c/o John Wiley & Sons, Ltd., The Atrium, Southern
`Gate, Chichester, West Sussex, UK, PO19 8SQ; Tel: 44(0) 1243 770 351;
`Fax: 44(0) 1243 770 432; E-mail: adsales@wiley.co.uk.
`Reprints: Reprint sales and inquiries should be directed to the Customer
`Service Dept., Attn: Gale Krouser, c/o John Wiley & Sons, Inc., 111 River
`Street, Hoboken, NJ 07030-5774; Tel: 201-748-8789;
`E-mail: gkrouser@wiley.com.
`Manuscripts: The Bell Labs Technical Journal welcomes submissions by
`membersof the Alcatel-Lucent technical community. The Journal issues
`calls for papers on selected theme issues through company communication
`vehicles. Persons with inquiries about submissions maycontact Jane M.
`DeHaven, Bell Labs Technical Journal, Alcatel-Lucent, Room 1D-426,
`600-700 Mountain Avenue, Murray Hill, NJ 07974; Tel: 908-582-6507;
`Fax: 908-582-3146; E-mail: j.dehaven@alcatel-lucent.com.
`Alcatel-Lucent trademarks and service marks are identified with the
`appropriate symbol uponfirst reference in each paperin this publication.
`Trademarksand service marks of other companies are identified with an
`asterisk after the mark and a note at the endofthe paper.
`Other Correspondence should be addressedto Bell Labs Technical
`Journal, Publisher, Professional/Trade Division, c/o John Wiley & Sons, Inc.,
`111 River Street, Hoboken, NJ 07030-5774.
`Bell Labs Technical Journalis indexed in Cambridge Scientific Abstracts
`(CSA/CIG); COMPENDEX(Elsevier); CompuScience Database (FIZ
`Karlsruhe); Computer & Communications Security Abstracts (Emerald);
`Computing Reviews (ACM); CSA Technology Research Database
`(CSA/CIG); Current Abstracts (EBSCO); Current Contents/Engineering,
`Computing & Technology (ThompsonISI); Current Index toStatistics
`(ASA/IMS); Ergonomics Abstracts ( T & F); Journal Citation Reports/
`Science Edition (Thomson ISI); SCOPUS(Elsevier); Web of Science
`(ThomsonISI).
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 002
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 002
`
`

`

`Bury Laks
`
`Technical Journal
`
`Volume 15, Number 1
`June 2010
`
`Alcatel-Lucent @
`
`® General Papers
`Guest Editor: Jack Kozik
`
`® Overview
`1
`Jack Kozik
`
`The Personal Internet
`3
`Koen Daenen,Bart Theeten, David Vanderfeesten, Bart Vrancken, Eric Waegeman, Jan Moons,
`Mihai Dragan, Monica Bojin, Mircea Strugaru, Knarig Arabshian, and Peter Bosch
`
`Application Creation for IMS Systems Through Macro-Enablers and Web2.0 Technologies
`23
`Anne Y. Lee
`
`@ A Social Network Service Solution Based on Mobile Subscriber Contact-Books
`53
`Xiang Feng and Qing A. Zhu
`
`Personalized Application Enablement by Web Session Analysis and Multisource UserProfiling
`67
`Armen Aghasaryan, Murali Kodialam,Sarit Mukherjee, Yann Toms, Christophe Senot,
`Stéphane Betgé-Brezetz, T. V. Lakshman, and Limin Wang
`
`# Utilizing a Personalization-Enabled Access Node in Support of Converged Cross-Domain
`Scoring and Advertising
`Andrey Kisel and Mark M. Clougherty
`
`@ On New Security Mechanismsfor Identity Management: Recognizing and Meeting Telecom
`Operator and Enterprise Needs
`Igor Faynberg, Mark A. Hartman, Hui-Lan Lu, and
`Douglas W. Varney
`
`Online Charging in the Roaming EPC/LTE Network
`115
`Yigang Cai and Xiang Yang Li
`
`Scalable Video Delivery Over MIMO OFDM Wireless Systems Using Joint PowerAllocation
`and AntennaSelection
`Shengjie Zhao, Mingli You, and Luoning Gui
`
`
`June 2010
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 003
`
`77
`
`95
`
`133
`
`(continued)
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 003
`
`

`

`
`
`® End-to-End Service Availability Support: Theory and Application
`Abhaya Asthana, Robert S. Blake, Richard D. Jordan, Reza Shafie-Khorasani, and Daryl J. Steen
`
`
`149
`
`® Delivering Quality of Experience in Multimedia Networks
`Harold Batteram, Gerard Damm, Amit Mukhopadhyay, Laurent Philippart,
`Rhodo Odysseos, and Carlos Urrutia-Valdés
`
`
`175
`
`@ Advanced Network Managementof the Optical Control Plane in Agile Photonic Networks
`Paolo Fogliata, ErmannoBelloli, Pietro V. Grandi, and Matteo Gumier
`
`
`195
`
`@ Facial Image Recognition Based on Fractal Image Encoding
`Xiutao Tang and Cuilu Qu
`
`
`209
`
`® Planning Energy-Efficient and Eco-Sustainable Telecommunications Networks
`H. Scott Matthews, Thomas B. Morawski, Amy L. Nagengast, Gerard P. O'Reilly,
`David D. Picklesimer, RaymondA. Sackett, and Paul P. Wu
`
`
`215
`
`Recent Alcatel-Lucent Patents
`
`237
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 004
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 004
`
`

`

`¢ Facial Image Recognition Based
`on Fractal Image Encoding
`Xiutao Tang and Cuilu Qu
`
`Fractal image encodingusesself-similarity to encode an image.Utilizing this
`feature, a new methodoffacial image recognition based onfractal image
`encoding is examined. Fractal image encoding can approximate any given
`image by capturing intrinsic self-similarity within the image. This method has
`someinvariance to rotation, scaling, translation, and luminance. The method
`encodes an imageof record through the use of an encoding dictionary
`composed of imagesin a face image database. The fractal codes are then
`placed ina file, and the fractal codesare iterated multiple timesto obtain
`decoding images. By comparing the peak signal-to-noise ratio of the image
`of record to all the decoding images, the image in the image database
`whose decoding image minimizes this norm is the recognized image. By
`experiment, the recognition performance achieved by this new method
`posted an average recognition rate of 94.54 percent on the publicly available
`database of faces hosted by the Cambridge University Computer Laboratory
`Digital Technology Group, whenfacial imagesat a side profile orientation
`are within 15 degreesof a full frontal image. © 2010 Alcatel-Lucent.
`
`
`
`Introduction
`a whole [3, 5]. To date, the global facial expression
`Facial image recognition relies significantly on the
`has proven to be of only limited use in facial image
`application and development of pattern recognition
`recognition.
`algorithms. At present, most of the research work in
`Teewoon Tan developed a face image recognition
`this area is only at the beginning stage. Facial image
`method based onfractal image encoding [4]. He first
`recognition presents manydifficulties, as described in
`proposed that an “image of record” could be matched
`[2]. For example, the face is not a rigid entity; it con-
`with another facial image in a face image database.
`tains complicated and abundant expression, so strict
`All facial images in the face image database were then
`feature matchingis ineffective. In addition, the data
`encoded, and the fractal codes were saved within the
`associated with a facial image is usually quite sub-
`database. For the process of recognition, fractal codes
`stantial, so storage requirementsfor input data can be
`wereiterated a single time to obtain decoding images,
`considerable. In fact, if the face pattern is unrestricted,
`and the decoding images were then compared to the
`input data may assume exponential proportions.
`neighbor distance of the image of record to obtain
`There are two methods by which a computer can
`image recognition. The main idea wastofilter high
`recognize a facial image: first, via local features and
`frequencyinformation such as changes of expression
`relationships such as eyes, nose, and mouth [8, 9],
`through fractal encoding andasingle iterative decoding,
`and second, bya statistical analysis of facial features as
`
`Bell Labs Technical Journal 15(1), 209-214 (2010) © 2010 Alcatel-Lucent. Published by Wiley Periodicals, Inc.
`Published online in Wiley InterScience (www.interscience.wiley.com) ¢ DOI: 10.1002/bltj.20433
`
`
`
`Alcatel-Lucent @
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 005
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 005
`
`

`

`
`and then to comparethe difference between the two
`
`Panel 1. Abbreviations, Acronyms, and Terms
`
`facial images. Although this method filters high fre-
`|FS—lIterative function system
`quency information byfractal transformation, the
`
`PSNR—Peak-signal-to-noise ratio
`
`recognition rate is not significantly improved.
`
`rms—Root mean square
`
`
`a new
`Utilizing the feature of self-similarity,
`
`methodof facial image recognition based onfractal
`
`image encoding is studied. This method encodes
`an imageofa face for the purposes of recognition though
`the use of an encoding dictionary composedof all
`images in the facial image database. Fractal codes for
`the imageare placedina file, and the codesare iterated
`multiple times to obtain decoding images. By compar-
`ing the peak signal-to-noise ratio (PSNR) of the image
`of record to all the decoding images, the image in the
`image database whose decoding image minimizesthis
`norm is the recognized image. By experiment, the
`recognition performanceof this new method achieved
`an average recognitionrate of 94.54 percent using the
`publicly available Cambridge University database of
`faces [1], whenthe face’s side profile image was within
`15 degrees of a full frontal image.
`
`2. The imageis divided into non-overlapping blocks
`of size Dax X Dinaye named domain blocks, usually
`De Bee
`3. Block D,, of size Dax * Dmax is transformed and
`classified into three major categories and 24 sub-
`categories by the gray value and the variance of
`block D;.
`4. Each rangeblock R;is classified into three major
`categories and 24 subcategories by the gray value
`and the variance of block R;. Next, search for the
`block D; which can match block R; best (usually
`there are manyD, blocks in the same class) from
`the same class of block R; and compute the rms
`between the best matching block D; and block R;.
`If the rms is smaller than the given value, record
`the related parameters.
`5. Otherwise, divide block R; by a quad tree, and
`return to step 3.
`6. End.
`
`¢
`
`«
`
`Image (a-2) is the decoding image that iter-
`ates fractal codes two times,
`
`Image (a-3) is the decoding image that iterates
`fractal codes three times,
`
`Principle of Fractal Image Encoding
`Throughresearch and analysis of a large number
`ol images, Wei, Shen, and Li have shownin [6] that
`there is self-similarity between a local image and an
`image as a whole, as well as self-similarity in different
`As shownin Figure 1, ImageAis the image of
`parts of an image. Fractal image encoding develops
`record. Image B and ImageC are other imagesin the
`aniterative function system(IFS) [7] by focusing on
`facial image database. Image B is the same person as
`the self-similarity in the image. The process of devel-
`Image A with a different posture. Image C is a differ-
`oping an IFS for the image begins by creating a sub-
`ent person from ImageA.
`block self-affine contraction to approximate another
`1.
`ImageA usesitself as the encoding dictionary and
`sub-block in the image. The decoding process utilizes the
`generates fractal codes.
`collage theorem to enable the IFS that can capture
`e
`Image (a-1) is the decoding image that iterates
`the imagetoiterate theinitial image multiple times to
`fractal codes one time,
`reconstruct an approximate image.
`To obtain the IFS of an image, we divide the
`image into non-overlapping range blocks. For each
`range block, we then search the domainblock that is
`most similar to the range block by affined transform
`and gray transform.
`
`e
`
`Image (a-4) is the decoding image that iter-
`ates fractal codes five times.
`
`Basic Algorithm of Fractal Image Encoding
`The fractal image encoding algorithmis as follows:
`1. The imageis divided into non-overlapping blocks
`of size Byyay X Bmax, named range blocks.
`
`2.
`
`Image A uses ImageB as the encoding dictionary
`and generates fractal codes.
`e
`Image (b-1) is the decoding image that iter-
`ates fractal codes one time,
`
`210
`
`Bell Labs Technical Journal
`
`DOI: 10.1002/bItj
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 006
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 006
`
`

`

`
`
`Table |.
`
`PSNR calculations.
`
`
`
`(a2)
`
`(a-3)
`
`(a-4)
`
`Image (c-4)/
`Image(b-4)/
`Image (a-4)/
`Image A
`Image A
`Image A
`
`17.738887
`25.437908
`34.215244
`
`PSNR—Peaksignal-to-noise ratio
`
`
`
`(b-2)
`
`(b-3)
`
`(b-4)
`
`We then compute the PSNR, as shownin Table I.
`The PSNR between image (a-4) and ImageA is 28.76,
`while the PSNR between image (b-4) and Image A is
`23.73. Meanwhile, the PSNR between image (c-4)
`(3)
`(c-2)
`and Image A is 17.59. ImageAis the image of record
`in the image database, while ImageB is obviously the
`“recognized” image. By using the PSNR, wecan rec-
`ognize images of the same person with different
`expressions and different postures.
`
`Figure 1.
`Facial image recognition based onfractal image
`encoding.
`
`¢
`
`e
`
`Image (b-2) is the decoding image that iter-
`ates fractal codes two times,
`
`Image (b-3) is the decoding image that iter-
`ates fractal codes three times,
`
`e
`
`Image (b-4) is the decoding image that iter-
`ates fractal codesfive times.
`Image A uses Image C as the encoding dictionary
`and generates fractal codes.
`°
`Image (c-1) is the decoding image that iter-
`ates fractal codes one time,
`Image (c-2) is the decoding image that iter-
`ates fractal codes two times,
`
`*
`
`3.
`
`e
`
`¢
`
`Image (c-3) is the decoding image that iter-
`ates fractal codes three times,
`2. The n “.dat” files are iterated five times by the
`Image (c-4) is the decoding image that iter-
`ates fractal codes five times.
`fractal image decoding program, which provides
`n decoding images.
`After five iterations of the same person withdif-
`3. The method of image recognition:
`ferent facial images—detailed in step 2 above, where
`The PSNR—thepeaksignal-to-noise ratio between
`Image A uses ImageBas the encoding dictionary—the
`the image of record and the decoding image—is the
`decoding image, image b-4, is very similar to Image A,
`quantitative measurement. The images whose PSNRis
`which is the imageof record. However, given the images
`higher than the threshold value are the recognized
`of two different persons—where Image A uses Image C
`images.If all PSNR values are lower than the thresh-
`as the encoding dictionary—the decoding image, image
`old value, thenthere is no image of the same person
`c-4,
`is unclear, and there are many small blocks in
`with an image of record in the face image database.
`the image.
`
`Face Image Recognition: Experiment Steps
`Cambridge University maintains a database of
`faces containing a set of facial
`images captured
`between April 1992 and April 1994 [1], which was
`used in the experiment. There are 400 facial images of
`40 persons with different expressions and different
`postures. Figure 2 provides a representative sample of
`images.
`Steps:
`1. The image of record is the input image, and the
`images in the image database are the encoding
`dictionary (assuming there are n images in the
`database). The image of record is encoded by an
`encoding dictionary composed of images in the
`image database, which then provides n “.dat”
`files.
`
`DOI: 10.1002/bItj
`
`Bell Labs Technical Journal
`
`211
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 007
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 007
`
`

`

`
`
`
`Figure 2.
`Imagesin the image database.
`
`2
`PSNR = POTogaeeee (1)
`:
`2
`(Sinn i<) i),
`mn
`
`Smnis the gray value of the (m, n) pixel of the image of
`record, S;,, is the gray value of the (#7, 7) pixel of the
`decoding image in the database, M is the height of
`the image, N is the width of the image.
`
`Experiment Results
`As shown in Table I, since the PSNR between
`image (a-24) and ImageA is the highest, image (a-24)is
`obviously the recognized image of Image A. That
`is, image (a-24) is the same person who appears in
`
`Image A with a different expression and different pos-
`ture. As shownin Table III, since the PSNR between
`image (b-8) and ImageB is the highest, image (b-8) is
`the recognized imageof Image B. Thatis, image (b-8)
`is the same person who appears in Image B, with a
`different expression and different posture.
`The time necessary for recognition is shown in
`Table IV.
`Facial image recognition based on fractal image
`encoding compares favorably to the 94.54 percent
`recognition rate for images within the Cambridge
`University databaseof faces, witha side profile facial
`orientation within 15 degrees of a full frontal image.
`
`Table Il.
`
`PSNR for Image A.
`
`
`
`
`
`
`
`PSNR with
`PSNR with
`PSNR with
`Decoding
`Decoding
`Decoding
`Image A
`image
`Image A
`image
`Image A
`image
`
`19.970701
`a-17
`20.264797
`a-1
`19.816535
`a-9
`
`19.452519
`a-18
`17.744489
`a-2
`22.508474
`a-10
`
`21.286626
`a-19
`20.531052
`a-3
`20.482979
`a-11
`
`19.818913
`a-20
`21.079812
`a-4
`20.190718
`
`17.211404
`a-21
`19.391354
`18.050826
`
`19.881216
`a-22
`20.618953
`16.788616
`
`21.518394
`21.392740
`a-23
`19:515317
`
`22.135498
`a-24
`23.953833
`18.573199
`
`PSNR—Peak signal-to-noise ratio
`
`212
`
`Bell Labs Technical Journal
`
`DOI: 10.1002/bitj
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 008
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 008
`
`

`

`
`
`
`
`Table Ill.
`
`PSNR for Image B.
`
`PSNR with
`PSNR with
`PSNR with
`Decoding
`Decoding
`Decoding
`Image B
`Image B
`image
`image
`image
`Image B
`21.834658 20.410737
`
`b-1
`18.700903
`b-9
`
`19.723254 19.514059
`b-2
`20.115239
`b-10
`
`b-11 21.77302720.556330 19.224120
`
`
`
`b-12 20.22677020.683712 19.582266
`
` 15.855981 19.715886
`
`20.066220
`b-13
`
`19.583121 19.370194
`19.420714
`b-14
`
`
`20.158694 20.067946
`19.739001
`b-15
`27.381136
`b-16
`19.386892
`20.486557
`
`
`
`
`
`
`
`
`
`
`
`
`PSNR—Peak signal-to-noise ratio
`
`Table IV. Time required for recognition.
`
`Numberof imagesin
`Time required for
`recognition (seconds)
`the image database
`50
`9.64 .
`
`100
`16.905
`
`
`33.421
`200
`
`300
`45.515
`
`400
`66.468
`
`Conclusions
`
`This paper studies a new methodoffacial image
`recognition based on fractal image encoding. The
`method takes an image of record and encodesit via an
`encoding dictionary composed of images in a face
`image database to obtain fractal codes, which are then
`placed in a file. We iterate the fractal codes multiple
`times to obtain a decoding image, and then compare
`the PSNR betweenthe imageof record and the decod-
`ing images. The experiment showsthat the method
`provides a favorable rate of recognition since it uses
`self-similarity within the facial image to eliminate the
`effects of expression and posture.
`
`Acknowledgements
`The authors would like to acknowledge Julian
`Cao, Roger Yu, and MannerLifor their contributions
`to this work.
`
`[4ined
`
`References
`[1] Cambridge University Computer Laboratory,
`Digital Technology Group, “Database of Faces,”
`<http://www.cl.cam.ac.uk/research/dtg/attarchive/
`facedatabase.html>.
`[2] G. Chen and F.-H. Qi, “Face Recognition Based
`on Fractal and Genetic Algorithms,” J. Infrared
`and Millimeter Waves, 19:5 (2000), 371-376.
`[3] A. Samal and P. A. Iyengar, “Automatic
`Recognition and Analysis of Human Faces and
`Racial Expressions: A Survey,” Pattern
`Recognition, 25:1 (1992), 65-77.
`T. Tan and H. Yan, “Object Recognition Based on
`Fractal Neighbor Distance,” Signal Process., 81:10
`(2001), 2105-2129.
`[5] M. Turk and A. Pentland, “Eigenfaces for
`Recognition,” J. Cognitive Neuroscience, 3:1
`(1991), 71-86.
`[6] H. Wei, L. Shen, and X.-H. Li, “Image
`Compression and Indexing Methods Based on
`Iterative Function System,” J. Image and
`Graphics, 7:11 (2002), 1198-1203.
`[7] W.-Q. Zeng, W.-Y. Wen, and W. Sun, Fractal
`Wavelet and Image Compression, Northeastern
`University Press, Shenyang, Ch., 2002.
`[8] J. Zhang, X. He, and J. Li, “Face Recognition
`Based on Geometrical Feature Points Extraction,”
`J. Infrared and Laser Engineering, 28:4 (1999),
`40-43.
`[9] J.-L. Zhou, Y. Zhang, H. Zhu, J. Guo, and J. Long,
`“A Study of Facial Feature Extraction Based on
`Dynamic Template,” J. Comput. Engineering,
`25:4 (1999), 53-76.
`
`DOI: 10.1002/bit}
`
`Bell Labs Technical Journal
`
`213
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 009
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 009
`
`

`

`
`
`(Manuscript approved November 2009)
`
`‘gi
`
`aa
`
`XIUTAO TANGis a memberoftechnical staff in the
`CDMAProduct Development Department at
`Alcatel-Lucent in Qingdao, China. His
`responsibilities include developing
`hardware such as base stations. He received
`
`
`
`a B.S. degree in industrial automation from
`the Harbin Institute of Technology in China and an M.S.
`degree in industrial engineering from the Beijing
`Institute of Technology. His research interests are
`mostly focused on codedivision multiple access (CDMA)
`and multi-user detection, CDMA network design and
`optimization, satellite network and transport protocols,
`and near field communication.
`
`
`
`CUILU QU is an instructor within the Department of
`Software Technology at Qingdao University
`in Qingdao, China. She received a B.S.
`degree in mobile communication from the
`Communications College of Zhangjiakou in
`China and an M.S. degree in signal and
`information processing from the Ocean University of
`China. Her research interests are mostly focused on
`computer and communications network security,
`network services and operation, fractal image
`compression, and image recognition. @
`
`214
`
`Bell Labs Technical Journal
`
`DOI: 10.1002/bitj
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 0010
`
`Google Exhibit 1014 - Google v. CSI
`IPR2025-00877 - Page 0010
`
`