Image Processing Reference
In-Depth Information
and a satisfaction threshold. A goal concept can also be reached by a “direct” strategy
if it is the first specified concept, or by an “indirect” strategy if it is specified within
the focusing rules. Depending on what strategies are used, the system's responses will
be different. We will now illustrate these comments with various experiments:
- direct analysis: the system directly searches for vehicles in the image, without
relying on context elements;
- indirect analysis: the system relies on context elements such as roads to search
for vehicles. Two focusing strategies are used in the examples mentioned: search for
vehicles “on” and “next to” roads.
10.6.1. Direct analysis
The first experiment consists of finding vehicles in an image using nothing but data
from the knowledge base attached to the vehicle concept. The analyzed image and the
associated knowledge base are shown in Figure 10.10.
{FIELD}
ROOT:IMAGES/SOURCE/Vol/vol.%02d.gif
EXTENSION:gif:BEGINNING:0:
NB:1:INCREMENT:1:
-------------------------- {GOAL}
CONCEPT:1: ------------------------
{VALIDATION} VAL:0.10
------------------------ {CONCEPT
1} NAME:VEHICLE: SIZE:1:
L:1:GREY:1:114:255:
L:1:HOMOG:3:1:10:
L:1:AREA:0:10:100:
D:1:DETECT:1:0:0
{CONCEPT 2} NAME:ROAD:SIZE:1:
L:1HOUGH:1:0.1:0.5:
L:1GREY:4:80:255:
L:1AREA:1:250:4000:
D:1DETECT:2:0:0
Figure 10.10. Direct analysis: the reference image is presented on the left
and the associated knowledge base is described on the right
In this image, two concepts are defined: road and vehicle. These concepts are char-
acterized by localization and detection methods. On the other hand, no focusing rule
is specified. Therefore, only the attributes related to the initial concept, namely “vehi-
cle”, are used. The models used in the localization phase to find the vehicles in the
image are statistical (gray levels and homogenity) and geometric (AREA). The popu-
lation of agents implemented in this example is shown in Figure 10.11.
Search WWH ::




Custom Search