Page 186 - Intelligent Communication Systems

CHAPTER 12/COMPUTER VISION 157
i 2.3.2 Laplacian Contour Detection
Whereas the main objective of space filtering is to eliminate noise through an inte-
gration operation, contour detection is performed through a differential operation
on the image. One of the procedures is as follows.
(1) The square of the difference between g(i +!,/') and g(i,j) and the square
of the difference between g(i,j) and g(i,j + 1) are added together.
(2) The square root of the value obtained in step (1) is assigned toflij). The
operation is shown in Eq. (9):

where g(i,j) is the gray-level value of pixel (i,J).
One drawback of this differential operation is that it emphasizes noise inclu-
ded in the image. To overcome this, a Laplacian operation is introduced. This cor-
responds to the secondary differential operation: f(i,j) is obtained via the term
{-g(i,j -l)-g(i- l,j) + 4g(iJ) - g(i + IJ)- g(ij + 1)}.
The Laplacian coefficient is given in Eq. (10)

By another method, (i, ) is obtained using the term
j
f
{-g(i - IJ- 1) - g(i- 1,7) - g(i - 1J + 1) -g(i - 1,7) + Sg(i,j)
~g(i+l,j)-g(i+lj-l)-g(i+l,j)~g(i+lj+l)}
The Laplacian coefficient is given in Eq. (11).

In both of these methods, the contour is determined. However, the direction
of contour cannot be obtained. To determine the direction of contour, methods such
as the Sobel operator are used.

12.3.3 Filters for Special Purposes
There are filters that detect specific shapes. The operation is the same as with tem-
plate matching.