for 
when the borders in the image posses higher rate to 
whole image, this method will be clumsy. As a 
result, we develop a new parallel algorithm; 
(1). For a digital image f, translate f for a pixel 
distance in E, S, W, and N directions, 
respectively, and then obtain a fmaw ( i — 
Be... 4) 
(2). For fran» take the logical negation operation, 
and then get feu (i=1,... ,4). 
(3). Holding the logical disjunction opreation to fai, 
obtain fpn. 
(4). Operating f and fay With logical conjuntion , 
finally get the borders. 
Obviously, this method is very efficient to be 
performed in parallel array processors. Its diagram is 
also simple(fig. 3). 
  
  
Fig. 3 Block Diagram for Border Following 
Fig. 4 is provided as a experimental result for border 
following. 
  
13 
  
Fig. 4 a). The Original Binary Image. 
b). The Border Image Found 
by Border Following 
4. REGION GROWING AND 
MATRIX REPRESENTATION 
A common image processing task is to separate out a 
particular region of the overall image on the basis of 
gray value or texture. The image is segmented into a 
feature region and a background region. The steps 
about region growing have been described previously. 
Meanwhile, the  proceduce is proved to be 
connectivity — preserving. Several new operations 
should be introduced, before presenting the block 
diagram of GROW. 
CTHRESH(£;,t)) (1,3) 
( , if the grav value=t at(i,j) 
0, otherwise 
Unless f(i,j) is undefined, in which case CABS(f)) 
(ij x. 
(DIV(f)I G,j) 
FG” if f(1,j) is real and not 
= , 
x, elsewhere