Biomedical Engineering Reference
In-Depth Information
The goal of tongue segmentation is to extract the tongue body from the origi-
nal tongue image, generally including the face, lips, teeth, etc. Only after accurate
results of tongue segmentation are achieved can the next steps be effectively con-
tinued. Some low-level image processing techniques have been applied in tongue
segmentation. Zhao et al. [74] presented a segmentation method based on math-
ematical morphology and the HIS (Hue, Intensity, and Saturation) color model.
Liu et al. [76] proposed an automatic approach for tongue segmentation based on
luminance information and the morphology features of the image. Sun et al. [77]
presented a method for tongue region segmentation based on a split-and-merge
algorithm.
Deformable models have already been adopted in tongue segmentation as
well. For example, Pang et al. [71] combined a bi-elliptical deformable template
with a snake model to segment a tongue image, where they introduced a novel term,
the
template force
, to take the place of internal force. In practice, unfortunately,
the methods mentioned above often fail to segment the tongue from out of its
surroundings. One of the reasons for this is that most of them do not take the color
property into account, which is an essential feature in such a task. In addition,
due to the pathological details on the surface of the tongue and the fragmental
weakness of the tongue's edge, tongue image segmentation based on deformable
models may converge to spurious boundaries.
In this section we propose an automatic segmentation scheme based on a color
GVF snake model that is applied directly on the original color tongue image. Note
that we do not carry out the segmentation using geometric deformable models
because the topology of the tongue is not very complicated and it is the unique
object of interest in the original image. Also note that we do not attempt to
establish a shape template to guide the convergence of the snake curves, like
the case in [16]. The shapes of tongue bodies captured in various diseases or
persons are quite different, so it is impossible to properly describe them all with
a predefined deformable template. Finally, to refine or correct the segmentation
result, an additional interactive segmentation technique is introduced.
Now we
will discuss the color GVF snake model.
3.2. Color GVF Snake Model
Although having extensive applications in the localization of object bound-
aries in bi-level and graylevel images, parametric deformable models have been
rarely used to find the boundaries of objects in a color image directly using color
information. Most researchers have employed the luminance characteristics of the
color images to drive the deformation processes, or applied deformable models
separately to some/each of the image planes (e.g., RGB [78] or other transformed
color spaces [79]), followed by certain combining strategies [80]. A review of
boundary localization methods based on parametric deformable models that make
use of the color data can be found in [81].
