Image Processing Reference
In-Depth Information
Fig. 17.5. ( Left ) A binary image with several regions. ( Right ) A connected component la-
belling where colors represent labels
L ( r i )= 0 ,
if
L ( r i )=0,
(17.13)
min k ( L ( r i−k ) G ( r k )) ,
otherwise ,
where the mask G for the backward scan (right-down) is given by
G = 0 0 1
111
(17.14)
The forward and the backward scans are alternately applied until no labels change in
the label image L .
Figure 17.5 shows a binary image consisting of the word “Direction”. Each letter
consists of one connected component except the letter “i” which consists of two.
Accordingly, the connected component labelling comes out with 11 regions, each
having its own unique label, shown as color of pixels. Using connected components,
one can apply further processing to individual regions, e.g., measuring their shape
properties, as we will discuss in what follows.
17.3 Elementary Shape Features
In practice, there exist a number of adhoc measures that may be sufficient to resolve
shape-related recognition issues. We list some of these descriptors in the following.
We assume that the image is f and that it contains one connected component
,
which is to be described w.r.t. its shape primitives. If there is more than one region
in the image, it is split up to several images so that there is no more than one re-
gion in each by connected component labelling. The points inside the region
D
D
are
Search WWH ::




Custom Search