synops nikhil

Embed Size (px)

Citation preview

  • 8/4/2019 synops nikhil

    1/9

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    Walchand College Of Engineering, Sangli.

    (An Autonomous Institute)

    Department of Electronics Engineering

    M.Tech. Part II

    Dissertation Phase I

    SYNOPSIS

    1. Name of Student : Mr. Mane Nikhil Niwas

    2. Name of Course : M.Tech. in Electronics Engineering

    3. Date of Registration : August, 2009.

    4. Name of Guide : Mr. B. G. Patil

    5. Proposed Title of Dissertation : Iris feature extraction and classification

    using FPGA.

    6. Synopsis of the Work :

    A) Problem Definition and RelevanceIris image processing having main four steps are image acquisition, image

    preprocessing, feature extraction and matching. Output of feature extraction

    block is nxm size image. To increase speed of matching stage need to reducedimension of template image without losing information. Singular value

    decomposition algorithm extract nxm size image to 1xm size array hencematching can be perform n times faster. Beauty of SVD algorithm is, it extracts

    maximum information at first element and minimum information at last elementin array hence we can ignore last elements for more optimization. Template can

    be matched with stored templates in read only memory in FPGA and classifyusing calculating hamming distance.

    This work deals with FPGA implementation of iris feature extraction

    using singular value decomposition (SVD) algorithm and iris pattern matching(i.e. classification) using hamming distance. Image data can be sent throughserial or parallel port of personal computer. Images will be taken from standarddatabase.

  • 8/4/2019 synops nikhil

    2/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    B) Present TheoriesIris biometric processing require large computation time than other

    biometric processing, so using single chip solutions for iris recognition system

    having constraint on speed and throughput. Now a days single chip solutions are provided by some manufacturer e.g. Texas instrument (TI) TMS320DM54xx,TMS320DM64x, OMAP series processors but not specific for iris image

    processing.

    Iris Recognition:Intelligent personal identification based on biometrics measurement such

    as facial feature, thermal emission, retina, iris, gait, voiceprint, gesture, palm- prints, fingerprints, handwritten signature, hand geometry etc. Iris is the only

    internal human organ that is visible from outside thus well protected from externalmodifiers. Fingerprint may suffer from harm and aging, voice patterns may be

    altered due to vocal diseases.The main steps in Iris recognition system consists of:

    y Image Acquisition:The first step of Iris Recognition is image acquisition. Normally, black

    and white camera should use for acquisition and must have enough resolution tocapture the details of the iris patterns. The illumination angle will determine the

    dark and light parts of the image. It is very important that one system implementsconsistent illumination, on the contrary the same iris may generate two different

    classes under two different illumination angles. Also, the pupil is an open door tothe retina, one of the most sensitive organs of our body, and extra care must be

    taken when shedding direct light over it.

    y Image Segmentation:The segmentation is process to remove non useful information, namely the

    pupil segment and the part outside the iris (sclera, eyelids, skin). Following are

    steps involve in segmentation. Detecting the Pupil Boundary

    Iris Edge Detection

    y Feature Extraction:Image is a huge redundant data. Single image contains lot of low

    frequency components within region of interest (ROI) hence feature extractionprocess extract valuable information from ROI into a small vector that representsunivocally the user. Feature extraction algorithm chosen is based on two

    requirements, performance of the whole system and complexity of the hardwarerequired to achieve it.

  • 8/4/2019 synops nikhil

    3/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    y Iris Pattern Matching:After the feature extraction the Iris templates are then matched with

    unknown templates using matching algorithms. Performance of this block should

    be the most robust part of the system. Hamming distance is simpler way to findout depth of matching. Matching and non-matching is based on hamming

    distance. If distance is minimum (i.e. below threshold value) pattern is match,otherwise corresponding Iris pattern is not in the database.

    7. Introduction:

    This work will do with combination of hardware and software. UsingFPGA speed, area, power and throughput can be optimize, also re-configurability

    obtained easily. Hardware system will be FPGA starter kit. The image data can beget via communication port of personal computer. Note that interfacing parallel

    port to FPGA kits needs extra hardware such as line buffers.

    Block diagram of system:-

    Figure 1: block diagram of system

    Personal computer:Acquisition of image is either from camera or standard image database. Todays

    cameras are able to provide compressed image (using JPEG standards) camera itself.Image segmentation can be done using Hough transform. These two steps can be

    performing with the help of MTLAB software. Segmented images are stored intopersonal computer due to limited memory in FPGA integrated circuit.

  • 8/4/2019 synops nikhil

    4/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    Serial / parallel communication:The segmented image data can be sent through communication port, to reduce

    data transmission time can use parallel interface, since transmission can be n times faster

    than serial interface. Where n is number of parallel bits (max 8).

    FPGA starter kit:

    Calculations of Singular Value Decomposition and hamming distance canimplements in FPGA kit using VHDL programming, size of S vector (template) is

    depends on architecture of FPGA kit.

    Single value decomposition algorithm:

    If X is a matrix with row and column dimensions n and m respectively, then n by

    n orthogonal matrix U and m by m orthogonal matrix V can be found such that,

    Where is a p by p diagonal matrix (p is the minimum of n and m). The diagonal

    elements of are the singular values of X and they are stored from largest to smallest.The above equation assumes that n >= m.

    Since U and V are orthogonal (and so their inverses are equal to their transpose),the above equation can also be written as:

    For large matrices, it can be impractical to compute U (which is n by n).

    However, U can be partitioned intoU = (U1, U2)

    Where U1 is n by p. ThenX = U1SV

    This factorization of X called singular value decomposition.The singular values in S are square roots of Eigen values from U or V. The

    singular values are always real numbers. If the matrix A is a real matrix, then U and V arealso real.

  • 8/4/2019 synops nikhil

    5/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    Flow chart for hardware implementation of SVD algorithm:

    Start

    Set communication parameters.

    Convert serial data to parallel vector.

    Combining parallel vectors becomes n x m size image

    A (n x m).

    Obtain transpose in matrix A as A (m x n).

    Obtain W1 (n x n) = A*A andW2 m x m = A*A.

    Obtain eigenvector of W1 is U (n x n) and W2

    is V (m x m).

    Obtain square root of eigenvalue of U or V is diagonal

    elements of S (p x p) matrix, arranged

    in descending form.

    Finish

    Store diagonal S (1 x p) matrix into temporary in on

    chip RAM

  • 8/4/2019 synops nikhil

    6/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    Flow chart for implementation of matching algorithm:

  • 8/4/2019 synops nikhil

    7/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    8. The Proposed Work:1. Study of SVD algorithm and matching algorithm.2. Study of FPGA implementation techniques for optimizing speed, power and

    performance.

    3. Study of Xilinx ISE design suite Software.4. VHDL codes and FPGA implementation using Xilinx software suite.5. Study the performance of system on the basis of area, power, throughput, etc.

    10.F

    acilities Available : Xilinx ISE Tool, FPGA starter kit, Library,Computer Lab, Internet etc

    11. Estimated Cost : 15,000/-(Approx.)

    12. Expected Date of Completion: July, 2011.

    Nikhil Niwas Mane Prof. Mr. B. G. Patil

    Student Guide

    H.O.D.

    Electronics DepartmentWalchand College of Engg., Sangli.

  • 8/4/2019 synops nikhil

    8/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    References:

    Papers:

    1) Liu-Jimenez, J. Sanchez-Reillo, and R. Sanchez-Avila, "FULL HARDWARESOLUTION FOR PROCESSING IRIS BIOMETRICS" security Technology, 2005.

    CCST '05. 39th Annual 2005.

    2) Christel-loc TISSE, Lionel MARTIN, Lionel TORRES and Michel ROBERT"Person identification technique using human iris recognition" France, July 2000.

    3) Li Ma, Yunhong Wang and Tieniu Tan "Iris Recognition Using CircularSymmetric Filters Pattern Recognition, 2002.

    4) Ignacio Bravo, Pedro Jimenez, Manuel Mazo, Jose Luis Lzaro and AlfredoGardel IMPLEMENTATIONIN FPGASOF JACOBIMETHOD TO SOLVE THE

    EIGENVALUE AND EIGENVECTOR PROBLEM , 2006.

    5) Li Ma, Tieniu Tan, Fellow, IEEE, Yunhong Wang, Member, IEEE, and DexinZhang "Efficient Iris Recognition by Characterizing Key Local Variations"IEEE

    TRANSACTIONS ON IMAGE PROCESSING, VOL. 13, NO. 6, JUNE 2004.

    6) Li Ma, Tieniu Tan, Senior Member, IEEE, Yunhong Wang, Member, IEEE, andDexin Zhang "Personal Identification Based on Iris Texture Analysis", IEEE

    TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,

    VOL. 25, NO. 12, DECEMBER 2003.

    7) Fernando E. Ortiz, John R. Humphrey, James P. Durbano and Dennis W. PratherA Study on the Design of Floating-Point Functions in FPGAs 2003.

    8) Javier Hormigo, Manuel Sanchez, Mario A. Gonzalez, Gerardo Bandera, JulioVillalba, Optimized FPGA Implementation of Trigonometric Functions with Large

    Input Argument Proc. Of XIX Conference on Design of Circuits and Integrated

    Systems, 2004

  • 8/4/2019 synops nikhil

    9/9

    Iris feature extraction and classification using FPGA.

    Department of Electronics Engineering,

    Walchand College of Engineering, Sangli.

    Websites:

    1. web.mit.edu2. www.xilinx.com3. wikipedia.org4. ieeexplore.ieee.org

    Books:

    1.Rafael Gonzalez, Richard E. Woods, Digital Image Processing, Pearson Education,India (2002).

    2.Xilinx vertex series user guide.3.Xilinx Spartan series user guide.