Image Databases for Face Recognition System - PowerPoint PPT Presentation

About This Presentation
Title:

Image Databases for Face Recognition System

Description:

Baltimore 34 NY Giants 7 (Jan. 28th, 2001) Attendance: 71,921 ... 'When tens of thousands of football fans packed into a Florida Stadium for Super ... – PowerPoint PPT presentation

Number of Views:49
Avg rating:3.0/5.0
Slides: 15
Provided by: YumikoShi4
Learn more at: http://web.cecs.pdx.edu
Category:

less

Transcript and Presenter's Notes

Title: Image Databases for Face Recognition System


1
Image Databases for Face Recognition System
  • Yumiko Shironouchi

2
Super Bowl XXXV 2000 Season
  • Baltimore 34 NY Giants 7 (Jan. 28th, 2001)
  • Attendance 71,921

3
Call It Super Bowl Face Scan I (Wired News, 2001)
  • When tens of thousands of football fans packed
    into a Florida Stadium for Super Bowl XXXV, they
    werent merely watching the game They were also
    being watched.
  • Face-Recognition software surreptitiously scanned
    everyone passing through turnstiles and flashed
    probably matches with the mugs of known criminals
    on the screens of a police control room.

4
Facial Scans
  • 3 processes of facial scan
  • feature extraction
  • search key creation
  • matching

5
Feature Vector
  • Three Main features of an image
  • Color histogram
  • Texture
  • Shape of object
  • It depends on applications which feature is
    extracted and converted into vector notations.
  • Images that have similar feature vectors they
    are similar images

6
Color Histogram
  • Vertical values represents the number of pixels
    that have the corresponding pixel value.
  • of pixel (value x)
  • total of pixels
  • one factor of feature vector (pixel value x)

0
255 (black)
(white)


(Bebis, 2001)
feature vector n(x 0)/total, n(x1)/total,
, n(x255)/total
7
Graph (shape of face)
  • Wavelet Transform
  • divide an image into high-frequency ingredient
    and low-frequency ingredient
  • extract of edges of object (face) analyzing
    low-frequency ingredient

upper original image lower
edge image (Looney, 2002)

8
Graph (cont.)
  • Pick up the feature points (eyes, nose and mouth)
    from the edge image to make a graph
  • Convert into a vector distance (or ratio to a
    unit distance ) to neighbor nodes and the angles
    between each edge

(Systems Biophysics, 2001)
9
For the efficient searching
  • Grouping images is necessary for faster search
  • Two access ways
  • - hashing (Grid Files)
  • - indexing (R-Tree)

feature vector of an image
10
Hashing
  • Grid File
  • Divide the space into grids arbitrary
  • Each grid becomes a key of searching

Image data
A grid represents a group of similar images
11
Indexing
  • R-tree
  • Grouping k (some positive integer) nearest images
    from a point (nearest k points search)

Above graph is shown in 2-dimensional, but
actually it is in multi-dimensional
12
  • representative vector
  • the center of feature vectors of images in the
    group
  • Groups of images are sorted and searched using
    the representative vectors.

13
Image Data Flow
Store or search
Grey arrow flow of the creation of image
database White arrow flow of the search of
similar images
Database
14
Reference
  • Systems Biophysics, the Institut für
    Neuroinformatik (INI), 2002
  • http//www.neuroinformatik.ruhr-uni-bochum.de/ini
    /top.html
  • Wired News, Lycos Inc., 2002
  • http//www.wired.com/
  • Dr. George Bebis, Associate Professor, Computer
    Science of University of Nevada, Reno
  • Dr. Carl Looney, Professor, Computer Science of
    University of Nevada, Reno
Write a Comment
User Comments (0)
About PowerShow.com