The purpose of preprocessing is to delete some image regions caused by noise or to fill some image regions to smooth the image pattern. Consider a point P(I,J) and its neighborhood in which we define :

(i) cross point: n_0, n_1....n_7

(ii)even cross point: n_0, n_2,n_4,n_6

(iii)odd cross point: n_1, n_3,n_5,n_7

Set :

(i) cross point: CP =

(ii)even cross point: CPe =

(iii)odd cross point: CPo =

The noise

suppression step considered here contains three parts :

(i) to delete the isolated point :(refer to Fig.2)

if P(I,J)=1 , and CPe=0 then set P(I,J)=0

(ii) to delete the dilation point :(refer to Fig.2)

if P(I,J)=1 , and CP=1 then set P(I,J)=0

(iii) to recover the missing point :(refer to Fig.2)

if P(I,J)=0 , and CPe=4 then set P(I,J)=1

Before moving to the action of thinning, we can "threshold" the image first. Generally speaking, there are two threshold methods : (i) bilevel thresholds; and, (ii) multi-level thresholds. These two methods divide the pixels of the whole image into two or more classes and increase the contrast of the image, so that we can deal with the following steps. During the process of thresholds, we use the bilevel threshold method{3}, so the original image is divided into two pixel classes; that is the below-threshold pixel and the above-threshold pixel. The threshold value "m" is expressed as the following equation :

m =

where ,

f : input image

n : total pixel no.

f(x,y) : the gray-level value of pixel(x,y)

When the threshold value m is calculated, then all below-threshold gray values in f(x,y) are replaced by z_0 and all above-threshold gray values are replaced by z_1 (z_0 < z_1). That is, after the threshold step, the whole image consists of pixels with only two gray values z_0 and z_1.

nextindexback