Research on Teeth Positioning Based on Wavelet Transform and Edge Detection - Advances in Image and Graphics Technologies - page 291

Through the vertical projection of the image after edge detection with the sum of
each list of pixels as the projection value, the projection curve of the teeth zone can be
obtained as shown in Figure 7 (a).

(a) Vertical Projection of Edge Image (b) Vertical Projection of Edge Image and Zone Average Value

Fig. 7. Edge Image Vertical Projection and Average Value

As shown in the vertical projection curve of the edge image, the projection value of
the teeth position is higher and that of the two sides of teeth is lower. Except for few
points, the projection value of the two sides is nearly zero. Therefore, the teeth zone
in the image can be obtained through threshold processing and then the boundaries
can be defined.
The author has designed a threshold calculation process, through which a valid
threshold value can be obtained. The calculation steps are as follow:

(1) Extract the maximum values in the edge vertical projection curve.
(2) Rank the maximums and define the zone of the top 10 maximums.
(3) Calculate the average of all non-maximums in the zone defined in Step (2).
Then this average is the threshold value.
The average value, i.e. the threshold value of the teeth position's vertical projection
can be obtained through the above steps. This average can reflect the basic feature of
the teeth position and effectively distinguish the teeth position and non-teeth position.
The experiment result in Figure 7 (b) shows that, both the left side of the first point
which is greater than the threshold value and the right side of the last point which is
greater than the threshold value are non-teeth zone; the zone between these two points
is exactly the teeth zone.

3 Experiment Result and Analysis

In order to verify the validity of the algorithm, the author has conducted experiments
on 50 X-ray images, among which 41 images can output accurate results. Thus the
accuracy rate of this algorithm reaches 82%. The experiment results demonstrate the
effectiveness of the proposed algorithm. The experiment results of part of the images
can be seen in Figure 8.