Docstoc

lee

Document Sample
lee Powered By Docstoc
					Normalizing illumination
Against varying light conditions
For face recognition
           Final Project
             ECE 738
           May 4th, 2005
                           Lee, Kang Eui
Introduction
   Face Recognition remains an unsolved
    problem in general
       Pose variation
       Illumination variation
   Canonical form approaches for illumination
    variation
       Normalize the variation in appearance
       HE, GIC, QIR
Histogram Equalization
          Result of HE
    Gamma Intensity Correction
   Correct the overall brightness of the face
    image to a predefined ‘canonical’ image
   Canonical image
       Face image under some normal lighting condition
   Correction is performed using Gamma Trans.



         Result of GIC
Regional HE(RHE) & GIC(RGIC)
   HE and GIC are global
    transforms
       Fail to compensate for the side
        lighting
       Localized transform is needed
   Ideally, partition should be done
    according to the structure of the
    facial organs.
       However, complicated region
        partition approach
       Instead, simply partition the face
        into four regions
Results of RHE & RGIC
  Regional HE     Regional GIC
 Quotient Illumination Relighting
I ij : image of ' object i' under the j - th lighting condition
I i 0 : image of ' object i' under canonical lighting condition
R j ( x, y ) : Quotient illuminati on
               1        N     I ij ( x, y )
R j ( x, y ) 
               N
                      I
                       i 1         ( x, y )
                                               , i  1,2,...,N , j  1,2,...,L
                               i0

                   I ij ( x, y )
I i 0 ( x, y ) 
                   R j ( x, y )
   Results




Result of QIR
Overall Result
using Eigenface approach
Used database: Yale Face Database B

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:5
posted:4/28/2011
language:English
pages:9