Page 237 - Mechatronics for Safety, Security and Dependability in a New Era

Ch45-I044963.fm Page 221 Tuesday, August 1, 2006 3:55 PM
3:55 PM
Page 221
Tuesday, August
Ch45-I044963.fm
1, 2006
221
221
An individual chromosome is represented by the arrayed bits whose length is same as the number of
edge pixels included in the template image. Each bit corresponds to each edge pixel in the template
image. The bit 1 shows that the corresponding edge pixel is used for matching, and the bit 0 shows
that the corresponding edge pixel is not used. Fig. 3 shows the evaluation method of an individual
fitness in the GA. The proposed method generates a new template table only the selected edge pixels
in the template image. Edge pixels are selected by an individual k whose fitness is evaluated by
matching results for learning images.
The fitness is evaluated by two parameters. The one parameter VS(k) means the ratio between the
maximum voted value and the second voted value. This parameter evaluates the reliability of
matching. The another parameter P(k) means the reduced ratio of the number of edge pixels for
matching. This parameter evaluates the possibility of high-speed matching. In the GA, the total
number of individual is set to 100, the continuation rate of all individuals is set to 50% and the
mutation rate is set to 1%.

EXPERIMENTAL RESULTS

(c)
Figure 4: Template images and template images with all edge pixels

Fig. 4 shows the four kinds of template images with gray scale and template images with all edge
pixels. The size of the objective images were 192x256 pixels. The experiments were executed by three
kinds of method described as follows in order to demonstrate the effectiveness of the pair of
curvatures as the matching key and the selection of edge pixels.
(Methodl) the matching method with the single curvature that was measured in a local area. (Using
the template image with all edge pixels)
(Method2) the matching method with the pair of curvatures. (Using the template image with all edge
pixels)
(Proposed method) the matching method with the pair of curvatures. (Using the template image with
selected edge pixels by GA)

TABLE 1
COMPARISON OF RECOGNITION RATE BY MATCHING KEY
Template (a) (b) (c) (d) Ave.
Methodl 84% 78% 76% 100% 84.5%
Method2 100% 94% 78% 100% 93.0%
The results of the methodl and the method2 were compared in order to evaluate effectiveness of the
pair of curvatures as the matching key. Table 2 shows the recognition rate that was obtained from the
results of matching by using 50 objective images for each template image. This result shows that the
superior result in the recognition rate was obtained by the method2 that was used the pair of
curvatures as the matching key.
Fig. 5 shows the generated template images with selected edge pixels by GA. The number of edge
pixels was reduced to about 28.1% on the average. It is expected that high-speed matching can be