Full text: Papers accepted on the basis of peer-review full manuscripts (Part A)

ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
The final color of the actual pixel is 
calculated so that the contributions of all suitable 
records are averaged. The suitable records are 
described in next section. 
Up to now only double resolution was 
considered. Also non-integer enhancement can be 
achieved. You have to change the initial resampling to 
get the basis for enhanced image the further procedure 
is analogis to described one. 
The further improvements of the method 
The records used in this method have 
arbitrary relative positions. So there can occur the case 
of irregular distributed records which degrades the 
resulting image quality. The reason for this is the use 
of arithmetic average by the determination of actual 
pixel color which cases that by irregular distribution 
of the records the unbalanced image is created. 
Under the term irregular distribution of the 
records we mean irregular distribution of the first 
detection object center coordinates of all records after 
the translation which consolides the integer parts of 
first detection objects. So of the records only that 
uniformly distributed in the plane are selected. The 
resulting image is determined only by such selected 
records labeled suitable records. 
The next undesirable efect of record 
averaging is the smoothing 1f edges and depression of 
resulting image objects contrast cased by bad linear 
properties and noise of the light-sensitive elements of 
the CCD camera. 
In effort of suppresing these defects there is 
pplied to resulting image folowing method which adds 
the information from first record to the result. This 
method is called image stabilization. 
The method of image stabilization: 
This method assumes that resulting image 
after reduction to records resulution must contain 
pixles with same color as on the records. The position 
of resulting image in choosed coordinate system is 
identical to first record position so the stabilization 
will use first record as reference. The pixel of first 
record covers exactly n° pixels of resulting image (n is 
a coeficient of integer resolution enhancement). Then 
must be valid that color computed by averaging of all 
resulting pixels covered by first record pixel equals to 
that first record pixel color. If resulting image doesn't 
fulfill these feature for some group of pixels the color 
of that group is modified to satisfy it. The color of 
pixels from this group are modified equally. If the 
color of one pixel from group is changed about one 
level, the further modification is possible after change 
of color of all pixels from that group. 
The tests showed that the stabilization 
reduced the effect of smoothing and reduction of 
contrast in resulting image. In some cases another 
effect of partial undersampling occured and grouping 
of pixles was observed in the resulting image. This 
undesitable effect was most visible on places where 
average color of pixel group was most different from 
corrensponding pixel color in first record. So the 
application implements this method wit possibility of 
user definable maximum level of color modification. 
The artifitially constructed records and results 
obtained by random shift method are in the Figure 10. 
As it is visible the stabilized image after thresholding 
differs from scene only by one missing line with half- 
pixle width. 
Il |l Il 
= 
  
object 
    
  
    
   
  
first record after thresholding 
after thresholding 
result after thresholding 
          
| p EE = 8 
stabilized image after thresholding 
Figure 10 Artificially constructed images and results 
Conclusion 
The paper described one of the methods for 
enhancement of viewed scene image. This method 
needed some additional modifications like image 
stabilization to get as best as possible results. During 
the tests images with 8-tuple resolution were obtained 
with quite nice results. This method can be further 
modified. Also as said before the result depends on the 
relative positions of records. One can construct the 
record series where the scene consists of regular black 
and white squares with square side size of one half of 
the record pixel. Whith composed such scene only 
images with one gray color can be recorded. From that 
information no improvement is possible when 
considering only translation of the record by the image 
snap. But if the records are snapped with rotated scene 
then it has more colors and the enhanced image can be 
A - 263 
 
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.