palm print recognition using local binary patt

Upload: ankur-gupta

Post on 04-Apr-2018

220 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    1/26

    Palm Print RecognitionUsing Local Binary Patternand Support Vector Machine

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    2/26

    1.INTRODUCTION

    The Key reasons for a Biometric PersonalIdentification

    Traditional Methods may allowunauthorized access.

    A password assigned to each user maybe guessed by an unauthorized user or

    forgotten by the authentic user.

    1/13/2011 INTRODUCTION

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    3/26

    1/13/2011 INTRODUCTION

    Fig. 1 System Block diagram

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    4/26

    1/13/2011 INPUT IMAGE

    Image from PolyU database(2nd version)

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    5/26

    1.IMAGE SEGMENTATION

    1.Algorithm to find points between the fingers2.Obtaining the Region of Interest3.Image Pre-processing

    1/13/2011 IMAGE SEGMENTATION

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    6/26

    1/13/2011 IMAGE SEGMANTATION

    II. A.

    Using a low pass Gaussian filter with a particularthreshold (say T), the palm print image isconverted from gray scale image to a binary

    image.

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    7/26

    Fig. 2 Key Point Detectionbetween 2 fingers.

    The set of 3 horizontal pixels just

    above the point of interest shouldbe at the binary value one. The set of three horizontal pixels

    just below the point of interestshould be at the binary value of

    zero. The pixels on the left and right of

    the points should have zero level. The pixels on the left and right of

    the points of interest at aparticular distance should havezero level.

    1/13/2011 IMAGE SEGMENTATION

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    8/26

    1/13/2011 IMAGE SEGMENTATION

    Fig. 3 (a) Palm Image before rotation (b) Palm Image just beforecropping the ROI

    II. B.

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    9/26

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    10/26

    1/13/2011 IMAGE SEGMENTATION

    Fig. 4 (a) Original Image (b) Square segmented ROI

    Using L2 as shown in Fig. 3(b) a square regionis obtained as shown in Fig. 4(b).

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    11/26

    Image Pre-Processing

    In order to increase the computational efficiency,the ROI is compressed using the waveletdecomposition.

    Furthermore, in order to enhance the main

    features in the palm such as the principle lines,ridges and wrinkles, a Laplacian filter is appliedto the ROI to sharpen the features of the palm,followed by histogram equalization to increase its

    contrast and finally a Gaussian filter is applied tothe ROI to smoothen out the edges.

    1/13/2011 IMAGE SEGMENTATION

    II. C.

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    12/26

    1.FEATURE EXTRACTION

    1.Sobel Operator2.Local Binary Pattern3.Feature Representation

    1/13/2011 FEATURE EXTRACTION

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    13/26

    Sobel Operator isused to find theedges of an image in

    a specified direction.The ROI is operatedon in eight directionsby the Sobeloperator

    1/13/2011 FEATURE EXTRACTION

    Fig. 5 Examples of directional responses

    derived using Sobel operator. (a) Originalpalm print images, (b) to (e) components ofthe images in the horizontal, vertical, positive45, and negative 45 directions.

    III. A.

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    14/26

    1/13/2011FEATURE EXTRACTION

    The Sobel masks used are given in Fig. 6.

    1 2 1 0 1 2 -1 0 1 -2 -1 0

    0 0 0 -1 0 1 -2 0 2 -1 0 1

    1 2 1 -2 -1 0 -1 0 1 0 1 2

    Fig. 6 3x3 Sobel Operators in different directions

    -1 -2 -1 0 -1 -2 1 0 -1 2 1 0

    0 0 0 1 0 -1 2 0 -2 1 0 -1

    1 2 1 2 1 0 1 0 -1 0 -1 -2

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    15/26

    Local Binary Pattern

    1/13/2011 FEATURE EXTRACTION

    III. B.

    The operator assigns a label to every pixel in animage by thresholding the eight neighborhoodpixels by the gray value of its centre. The featurevector is formed by concatenating thethresholded binary bits in the anti-clockwisedirection, which is shown in Fig. 7.

    An LBP string is called uniform if it consists of atmost 2 bit-wise transitions from 0 to 1 or vice-versa.

    This way the operation results in a set of 59 LBPpatterns denoted by {1, 2 59}.

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    16/26

    1/13/2011 FEATURE EXTRACTION

    Fig. 7 Obtaining a Binary label for a pixel

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    17/26

    Feature Representation

    1/13/2011 FEATURE EXTRACTION

    III. C.

    A histogram of labels can be constructed as:

    The palm print images are divided into 9 equallocal regions and the texture descriptor is extracted

    from each region independently. Application of the Sobel operator in four directions

    yields four Sobel images which are divided into 9sub-windows each being a 40x40-pixel image.

    Thus, the texture descriptor for a given palm printwill have a size of {59 (no. of labels) x 9 (no. ofsub-windows) x 8 (direction components)} asshown in fig. 8.

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    18/26

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    19/26

    MATCHING

    Chi-Square Method Support vector Machine

    1/13/2011 MATCHING

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    20/26

    Chi-Square Matching

    Where Sb and Mb are normalized enhanced

    histograms to be compared; indices b refers to the bthbin of the histogram, which in this case varies from 1to 59.

    1/13/2011 MATCHING

    The Chi-Square distance between the twohistograms S and M can be defined as

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    21/26

    1/13/2011

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    22/26

    Support Vector Machine

    SVM looks for Hyper Plane Maximizes Margins Minimizes quantity proportional to number

    of misclassification errors Creates decision boundary defined by:

    1/13/2011 MATCHING

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    23/26

    A few kernel function that have been used inthis study are Radial basis function : Sigmoid kernel :

    Polynomial kernel function: Linear kernel:

    1/13/2011 MATCHING

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    24/26

    The experiments were performed on 2different sets of Databaseo PolyU_Palmprint_Database (2nd Version): It

    contains 40 palm images of , 193 differentuserso IIT Delhi Database : This database was built at

    Biometrics laboratory, IIT Delhi and contains a

    minimum of 7 palm-images of 109 users

    1/13/2011 IMPLEMENTATION &RESULTS

    IMPLEMENTATION

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    25/26

    1/13/2011 IMPLEMENTATION &RESULTS

    Kernel Function used withSVM

    PolyU Database IIT-D Database

    Linear 0.50 1.00

    Polynomial 1.00 1.60

    Database Chi-SquareMatching(EER %)

    GAR whenFAR=1%

    GAR whenFAR=2%

    PolyU 2.11 95.72 97.72

    IIT-D 3.51 92.23 94.45

    Table 1: Error Rate plot using SVM as matching technique and different kernelfunctions.

    Table 2: Results using Chi-square matching technique

    RESULTS

  • 7/29/2019 Palm Print Recognition Using Local Binary Patt

    26/26

    1/13/2011

    Thank You

    Suggestion Please.