j-th column after stretching, f(i, j) is the gray value of the 
pixel after removing the border lines, fy (i,j) is the gray 
value of the pixel in the background image, ave, is the 
average gray value of the original image, fy. and foi 
are the maximum and minimum gray values of the 
background image, and the value range of k is [1, 2]. When 
fpr (ij) = fprmax>» the degree of stretching is greatest, and 
when fi, (i,j) = fbkmin> the degree of stretching is least. 
(3)Adjusting the average gray value of the image 
In order to keep the original image’s average lightness, adjust 
the gray value of the image after stretching. The formula is as 
follows: 
f'(,j) 2 f (1j) * aveg; — ave (9) 
Where f'(i,j) is the gray value of the pixel in the final 
resulting image, and ave is the average gray value of the 
image after stretching. 
4. EXPERIMENT AND ANALYSIS 
We chose many aerial images to validate the improved 
algorithm. The result showed that the improved algorithm 
can get a better dodging result than the original one, and can 
make the contrast of the whole image more consistent. In 
order to analyze the algorithm in greater detail, here we 
select one aerial image to show the dodging effect. 
  
Figure 3. The original image 
It can be seen from the original image (Figure 3) that the 
center region is brighter and the surrounding region is darker. 
In addition, it has a higher contrast and it is clearer in the 
bright area. 
We divide the original image into blocks and acquire the 
background image using the Gaussian low-pass filter with 
the same cut-off frequency in every image block. Figure 4 
shows the resulting images when Dg — 2 and D, = 4. 
  
Ed 
(b)The resulting image when D, — 4 
  
Figure 4. The resulting image when Dj is the same 
It can be seen that for the regions selected by red frames, the 
dodging effect is more satisfactory when D, = 4. However, 
when Dy = 4, there are black shadows around the houses in 
the region selected by the blue frame, which influences the 
image quality. 
In order to solve this problem, adopt the improved method 
put forward in this article and divide the image into blocks 
when producing the background image. Here we divide the 
image into four overlapping blocks and set the value range of 
the cut-off frequency to [2,4]. The resulting image is as 
follows (Figure 5): 
    
  
  
  
      
    
     
       
      
      
  
    
  
   
   
Fig 
Through 
is the m: 
minimun 
will not 
the min 
maximul 
dodging 
In order 
use som 
resulting 
is to firs 
the lowe 
then con 
be differ 
For exar 
the low 
lightness 
mean an 
dodging 
Therefor 
several s 
manually 
variance 
that mea