S are suitable for 
lor infrared aerial 
. Since the Gray- 
> Shades of Gray 
and propose this 
iges, and analyze 
rial images. After 
ED ON COLOR 
depicted here as 
of color 
hadow removal 
affects the effect 
1adow detection 
, which requires 
ell as a digital 
s not easy to get 
etropolitan area, 
cteristics in the 
ormation except 
nore commonly. 
but this is not 
ng images; it is 
1ges. 
In this paper, we use a shadow detection method that based on 
the image grey characteristics, which does not need the 
geometric information or DSM. Under the analysis and test 
result, the specthem ratio and Otsu threshold segmentation 
methods is applied to detect building shadows on urban aerial 
true color and color infrared aerial images in this paper. 
Segmentation is done in HSI space (H-hue, S-saturation, I- 
intensity), where the building shadows are detected and the 
impact of grass land and trees is eliminated based on 
mathematical morphology. Shadows in urban aerial image are 
detected by this method: 
First, transform the color image from RGB space to HSI space; 
from this the intensity component (T) and the hue component (H) 
are obtained. 
Second, get the specthem ratio image (( H + 1 )/ (+ 1 )) ; the 
grey value of the shadow region is larger than non-shadowed 
region in the ratio image. 
Third, apply Otsu method to determine the segmentation 
threshold for the ratio image; the image is then segmented and 
the candidate shadow region image is obtained. 
Lastly, extract the building shadow in the candidate shadow 
region image (the result from the third step) based on the 
morphological differences between building shadows and trees. 
The segmentation image is filtered by median filter in order to 
remove noise first; the resulting image is then processed by 
morphological erosion and dilation operators to get the shadow 
region. 
By shadow detection, the aerial image is distinguished as 
shadowed regions and non-shadowed one. 
3.2 Shadow removal 
After shadow detection, the aerial image is divided into two 
parts, a shadowed region and a non-shadowed region. The light 
source color is determined based on this result. The light source 
color value is calculated and the shadow is removed in the band 
image of the original aerial color image. The step of the shadow 
removal is as follows: 
(According to the shadow detection result, the band i of the 
original aerial color image s,(x,y) is divided into a shadow 
region image g,(x, y) and a non-shadowed one h,(x, y). 
S, (x, y) 7» g, Gc y) U A (x, y) (3) 
(2 The source light color of the shadow and non-shadowed 
regions (e,,e,) are calculated by equation (2) based on Color 
  
  
Constancy. 
[ 1 
(33 (a (xm) Y 
a FRE NE 
3 V 
os Ge »)) y (3b) 
os Th Fe 
C \ 7 
  
For e, ; the xy-summation is on the shadow region, M N is the 
pixel numbers on the shadow region. For €, ;, the xy- 
summation is on the non-shadow region, M N is the pixel 
    
       
    
   
    
   
  
  
    
    
   
   
    
  
   
    
     
   
   
    
    
     
  
   
   
    
  
  
     
  
     
  
   
  
  
  
   
   
    
    
numbers on the non-shadow region. K ,is scale factor, the 
(e, ;,e; ;) defined the k,, K.., k, in shadow regions, i means 
band i. 
(®) Changing shadow region light conditions to the standard 
illumination condition, as follows: 
8_b,(x,y)= g,(x y)/e, ; (4) 
Likewise, the standard illumination condition of non-shadowed 
region is calculated. 
h bj(x, y) = h(x, Ve, (5) 
After the above processing g_b,(x,y) has the same 
illumination condition as A b,(x, y). Then the illumination 
condition of shadow region is like equation (4). 
g_h(x,p)=g(x,y) e, ‚fe, (6) 
Finally, the shadow removal image is obtained as follow: 
new_S,(x,y)= g_ h,(X, y) U h; (x, y) (7) 
4. EVALUATION OF AERIAL IMAGE SHADOW 
REMOVAL RESULT 
Visual evaluation is a commonly used approach for quality 
assessment in the shadow removal. However, in this paper some 
quantitative evaluations are utilized to assess quality based on 
statistics. Here, statistical characteristic indices include 
brightness, contrast and average gradients. 
Brightness is calculated as equation (8) 
M N 
2,2, 5,66») 
Brightness, 5121  — — (8) 
MxN 
To reflect the black and white contrast of image, the contrast is 
calculated as follows: 
  
M N 
M SG y)- Brightness, | 
Contrast, — A MIN (9) 
x 
  
Average gradient reflect the amount of image detail. The 
calculating formula is: 
  
: 1 uan [Lae ALT) »/ (10) 
CC OMF DON -1) > Je A rt dy 72 
Where i means the band 7 of the image. 
The shadow removal results have been compared by the three 
indices.