Information Technology Reference
In-Depth Information
Fig. 3. Segmentation results (The black dots are the initial shape boundary and the
white dots are the final shape boundary).
the thickness of the parabola. For the simplicity of the problem, we have not yet
considered the axes rotation. Initially, we assumed symmetry in the tongue shape
between the left and the right portions of the tongue. Then we accounted for
more flexibility in the tongue shape by having different values of a , θ ( θ 1 left and
θ 1 right ), α ( α 1 left and α 1 right ) and different parabola equations for the left and the
right portions. Further, two additional circular portions were added to the left
and the right portions giving different values for θ ( θ 2 left and θ 2 right ), α ( α 2 left
and α 2 right )and r . The template is given a good initial guess point ( x 0 ,y 0 ).
All the parameters are varied within a certain feasible limit (which defines the
range of shapes the tongue template can take). The energy of the points ( E )is
calculated for all the possible templates. The average E is maximized and the
resultant vector (or template corresponding to this maximum E ) is the desired
solution.
The algorithm was implemented in Matlab TM . The black line indicates the
initial starting template and the white line gives the final solution. Result sam-
ples are shown in Fig. 3.
Currently, we provide just a single point for the initial point (i.e. ( x 0 ,y 0 )
or the lowest point of the tongue). It is desirable to provide more initial points
that can be located visually. A total of 5 points on the tongue could fix the
weakness mentioned above. Further, more constraints and finer step variations
on the parameters can give more accurate results. Using an array of templates
(instead of just a single template) can also give better results.
5 Feature Descriptions
Features on the tongue include color and texture. Most TCM practitioners do
not have numerical descriptions of the color or texture features. Instead, they use
analogies or qualitative descriptions, such as 'network-like cracks', or 'sandpaper-
like surface'. It is possible to develop a scheme to map the qualitative descriptions
to fuzzy sets of feature values.
However, in this chapter, we only discuss the numerical expressions of the
features. In this study, we use CIE L*a*b* color space to represent the color
features on the tongue. We also use several texture analysis methods that can
correspond to human-descriptions of textures: 1) calculating fractal dimension
Search WWH ::




Custom Search