Page 196 - Intelligent Communication Systems

CHAPTER 12/COMPUTER VISION 167

7X1, 1) 7X2, i) TXJV, i)
r(i,2) r(2, 2) T(N, 2)

. .

7X1, AO 7X2, AO T(N,N)

FIGURE 12.13 Area segmentation.

for each small area, a homogeneous larger area is created by integrating the small
areas that satisfy the following:

Conventionally, 3 x 3 or 5 x 5 small areas are used for integration.
Another method creates a homogeneous larger area by using a power spec-
tram of a small area. Given the power spectrum Gy of a region with NxNpixels,
G is as defined in Eq. 45:

where G is a feature vector and shows a texture in the area (OQ-fcfc). Given a feature
vector G in each area, the distance between areas k and / is as defined in Eq. (46):

where G k, G t are feature vectors of areas k and /, respectively.

12.4.2 image Compression

Image compression is useful for image analysis, image storage, and image trans-
mission. The compression is accomplished by using a pyramid data structure or a
tree structure.
12.4.2.1 Method Using a Pyramid Data Structure
2n
A series of images whose original image size is l/(2 ) (n = 1, 2, ...) is called a
pyramid data structure. As shown in Figure 12.14, in the case of a binary image,
the value of a pixel on a higher level is determined by using the conjunction of the
values of the four pixels on the lower level. For example, if at least one pixel with