| pho- 
rreihe 
rsität A Rule-Based Expert System for High-Quality 
Image Enhancement 
|Ohy- 
jets. Zhang Xiangqian, Li Xiaofeng, and Yu Xuchu 
‚ätica Zhengzhou Institute of Surveying and Mapping, Henan, China 
lona, 
ABSTRACT: This paper presents a rule-based expert system for the enhancement of 
tech- image contrast. Firstly, an input image is divided into several block subimages and 
ybrid the mean and variance transformation of each subimage are applied based on such 
Pho- rules. Secondly,based on the Gaussian overlay image recombination scheme, the en- 
hanced subimages are recombined together into a high-quality output image which 
ley will be devoid of defect where the subimages are connected. This paper puts 
rétrie emphases on the discussion of the mean and variance transformation, the 
al on establishment and modification of the rule base, and the localized enhancement tech- 
PRS, nique of the block subimages. The experimental results demonstrate that the 
rule-based method is superior to most traditional ones. 
KEY WORDS: Expert System, Image Quality, Enhancement 
t by 
Engi- owadays, there are many algorithms for con- enhancing image. The mathematical expressions are il- 
N ins enhancement. In the use of them, the lustrated as follows. 
pal, common problem faced is how to determine Let Is(1,)) Ms and Vs indicate the grey level of a 
ng of parameters in tranformation. Different selection of pixel at the position(i,j),mean and variance of the 
DTM parameters will result in defferent enhancement ef- source image Is respectively. Io(i,j), Mo and Vo indi- 
ional fects, and therefore the selection of parameters de- cate the grey level of a pixel at (i,j), mean and 
0. IV. pends on the properties of image to be processed. varcance of expected output image Io respectively, 
| In the past, the method of "probing" was used for where i=1,2,- M, j=1,2,:- N. 
nom. the selection of parameters in image processing. As Suppose Io(i,j) = A-Is (i,j)+ B (1) 
it is troublesome and short of memory function, betawse Mog YYLGj) Q) 
k (at the "probing" procedure is always repeated in pro- MN i-i 
mag- cessing images of the same properties. Ve = a Y Au (.j) -M2 (3) 
In this paper a rule-based expert system for con- MN 1j 
trast enhancement is presented. An initial rule base then: 
Wkers. is constructed by the image processing Knowleise of Mim ds WS LDTAM,H (4) 
man on the basis of the mean and variance MN =i 
lly & transformation. After that, the initial rule base is vos 3550 MP ARM (5) 
modified according to the expert evaluation on the MN ie 
experiment results for images of various properties. from(4) and(5), we get 
aches Thus, a rule base of expert knowledge is formed. In Av VIV, , B=M=M, J Vo/ Vs 
dd practice, repetition of probing for the selection of then (1) can be rewrited as 
pod parameters is avoided and an image can be en- FR 
hanced as expert expects by using rule-based con- IC DV hd) + Mo: Msy Vo/Vs (6) 
trast enhancement. 
oduc- In order to enhance the image which is strongly equation (6) is transformation relation between the 
Man- dependent on the spatial variation of scene grey level of input image Is and that of output im- 
Bul- illumination, the establishment of a rule base for lo- age Io. The transformation function curve is illus- 
calized enhancement and the method of eliminating trated in Figl. 
soft- the border effect resulted from localized o 
enhancement between block subimages are also dis- 255 
cussed in this paper. li -Ms-M,. 
GS MEAN AND VARIANCE TRANSFORMATION GU E AL 
Mean and variance transformation is to transform 
the mean and variance of a source image into those Figl. 
Is 
of an expected image so as to attain the goal of 
  
665