GEOMETRIC MODELING USING POINT CLOUDS 29910.5 Planar Contour Models
MRI and CT both yield object boundary information as data points placed in parallel planes or slices,
which can be arranged as contours to represent a 3-D object. In medical applications, computed
tomography is used to scan the interior of the body. It is effective for capturing images of bones and
dense organs such as the brain and abdominal region. CT scanners produce images by firing X-rays
at the region of interest and measuring the intensity of the rays after they have passed through the
body. Industrial CT scanners can be used for image engineering of artifacts to detect cracks or holes.
MRI is effective for producing images of soft tissues and is especially useful for detecting tumors in
the human body. A set of 2-D CT or MRI scans is treated as a single 3-D image. Usually, the images
produced by CT and MRI scanners are noisy, and in medical applications they capture densities for
more organs than are needed for the study. These images must be filtered to reduce noise, and further
must be thresholded and segmented to isolate regions or organs of interest. CT and MRI scans are
available in a variety of vendor-specific formats. For medical images, the American College of
Radiology and the National Electronics Manufacturers Association have set a standard called Digital
Imaging and Communications in Medicine (DICOM). For displaying and manufacturing these medical
models, we need to develop a mesh of facets across contours to represent the bounding surface of the
object.
10.5.1 Points to Contour Models
CT or MRI scanning yields a series of cross-sectional intensity images. Each such 2-D image is
composed of pixels. A pixel with value 1 (a black pixel) represents void while that with value 0 (a
white pixel) represents material. Pixels may also have values between 0 and 1 in the grey range. First,
planar contours are constructed from the data contained within a 2-D slice of the 3-D image. In each
2-D slice, there are one or more material blobs. The edges of those material blobs are located and
from them an ordered list of points is formed. If the points are connected with straight-line segments,
we obtain a polygonal contour representing the cross-section of the object. Two stages are involved
in extracting contours from a 2-D image. The first stage, called component labeling involves labeling
all blobs in the image. The second stage, called edge following, requires to follow the edge of each
blob and form a list of points describing the contour.
A stepwise illustration of the contour extraction process is as follows:
Initially, the image quality is improved by removing the noise. These images may be cleaned using
linear and non-linear filtering techniques applied to several kinds of noise. A Mexican hat low pass
band filter function is shown in Figure 10.3(a) and an example illustration of noise filtering is shown
in Figure 10.3(b). Once noise is filtered, the grey scale image is converted to an intensity image using
threshold. An appropriate threshold value or a range is chosen and all pixels above this threshold or
in the range are flagged as 1, otherwise as 0. For Ith as the intensity threshold chosen, the intensity
Figure 10.3 An example image processed for noise removal(a) Mexican hat low pass band filter (b) Image before and after noise filtering