| Summary: | 碩士 === 國立中正大學 === 電機工程研究所 === 82 === In many applications of computer vision and medical imaging,
three-dimensional (3-D) image is usually generated from a
sequence of two-dimensional (2-D) cross-sectional images. In
this research, we developed a 3-D reconstruction system which
resolved the following two major problems: (1) 2-D image
segmentation, (2) 3-D image reconstruction. For 2-D image
segmentation, a new method which is suitable for separating
human organs is proposed. Given a set of CT images, our method
can segment next consecutive image automatically using the
information of the last segmented image. The user''s
intervention can be minimized. The segmented image still
preserves the characteristic of texture, offers more
information for 3-D reconstruction and volume rendering. A new
reconstruction approach is also proposed. Not only the surface
but the whole organ is reconstructed. Past approaches are
mostly dedicated in the surface reconstruction. It is difficult
for further 3-D manipulation like dissection or slicing from
arbitrary angles. We improve the gray-level interpolation and
take the advantage of shape-based interpolation to satisfy our
goal. The whole reconstruction is proceeded with the following
three steps. First, correspondence is established between the
points in the consecutive images we want to interpolate.
Second, apply median filter to adjust the incorrect
correspondence. Finally, use this correspondence to estimate
the data between the consecutive images by linear
interpolation. By the progress of technology, the graphic
capability of computer is more and more powerful. Incorporating
with some special instrument the applications of virtual
realities like surgery simulation can be easily implemented. In
these applications the volume data is more important than
surface data. Hence our system establishes an important basis
for future medical imaging applications.
|
|---|
