ng 
lot 
d to 
N- 
/ 
Xf 
ess 
the 
S 
Number of pixels 
A rejected underexposed X-ray with a high level of noise 
was selected (figure 2). The pre-processed histogram 
(figure 3) showed a low-radiance, low-contrast scene and a 
bi-mode image. In fact, the original image was so 
underexposed it was totally invisible without contrast 
enhancement. Since the image is a bi-mode, a piecewise 
equalizer was used to enhance it. The gray levels between 
5 and 14 were stretched to between 50 and 255 while the 
rest of the image was not changed. Then the process was 
reversed and the gray levels between 15 and 30 were 
stretched to between 50 and 255. The two obtained 
enhanced images are shown in figures 4 and 5. By 
analyzing the two figures, it can be seen that the noise is the 
dominant feature in the image. The knee-cap and the tibia 
bone were almost totally lost. 
F que 2. Histogram equalization of the 
original X-ray. 
40000 
30000 
20000 
10000 
  
0 10 20 30 40 50 
Gray level 
Figure 3. Histogram of the original image 
  
    
Figure 4. Piecewise equalized image 
for the gray values 5 to 14. 
  
     
   
  
    
   
  
   
  
   
  
  
  
   
   
    
  
    
   
  
  
   
  
  
   
     
  
Figure 5. Piecewise equalized image 
for the gray values 15 to 30. 
Low pass, median, and mode filters were applied to the 
image at different kernel sizes from 3*3 to 9*9, At each 
kernel size the smoothed image was compared both 
visually and statistically with the original image and its 
histogram. 
Figures 6, 7, 8, 9, 10, and 11 show the obtained histogram 
and the resultant image for 3*3 and 5*5 kernel size. It 
should be emphasized that all filters were applied to the 
original image prior to any enhancement because existing 
noise would be enhanced. 
  
Figure 6. Spatial domain low pass filter 5*5. 
40000 
30000 
20000 
Number of pixels 
10000 
  
0 
0 10 30 40 50 
20 
Gray level 
Figure 7. Histogram of low pass filter 5*5 at 6.