The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B6b. Beijing 2008
horizontal detail coefficients (LH), the vertical detail
coefficients (HL), and the diagonal detail coefficients (HH) [10] .
cA
row
*j Lo_D
HIJD
column
„3' 42Ïr\ cAj,,
column
Hi_D "j—4a"T» \ —cD%,
column
*■ Lo D | 2J1 ■
J“~*i 2 * i U co 1 umn
- Hi_o 4.2 if'].. .
(a)
£vt /*i—*£* Î 2 }
column
c! *>■' —*1 1 Î 2 \~
•c
Lo_R
V
column
*
MI_R
r
co 1. umn
■T
Lo_R
L
column
c
HL.*
>
Lo R j
H1_R
-cA.
——« 1 J2
Figure 2. Diagram of 2-D WT decomposition (a) and
reconstruction (b)
Figure 5. Emulational image of non uniform illumination model
We can see that approximate and detail coefficients of the even
illumination image are almost in a plane, and the uneven
illumination image detail coefficients are almost in a plane,
only the extent enlarged, but the approximate coefficients is not
in a plane at all.
J* ¿“p J ;
1 level WT
LL
HL
LH
HH
Figure 3. Diagram of Single-level 2-D wavelet transform
The attenuating illumination model represented in formula (8)
shows that the illumination attenuates slowly from the image
central point to the around. This means that, the illumination
attenuating model corresponding to the low frequency part. In
wavelet transform, the low frequency signal is correspond to the
approximate coefficients. So it can be concluded that the
approximate coefficients are most sensitivity to the illumination
attenuation. While the detail coefficients correspond to the
detail information such as image edges.
For testing the above feature, emulational image of even
illumination image is given in Figure 4, illumination non
4
uniform model accord to cos ® is given in Figure 5. After one
level wavelet transform of these two figures, wavelet
coefficients are shown from Figure 6 to Figure 9.