Next: 5.3.2 Real Scans Up: 5.2 Methods Previous: 5.2 Methods

5.3.1 Point Simulations

For the point simulations, m points were randomly generated in 3-D space within spheres of radius r or shells with outer radius and thickness of 5 mm to produce the target configuration, A. A is randomly translated and rotated (uniform distribution) about the center of the specified spherically symmetric boundary within mm and degrees, respectively, to produce an exactly corresponding but un-registered configuration, B. The un-registered configuration with homogeneous homology error which is to be transformed for registration, , is produced by perturbing the 3-D position of each point of B within homogeneous Gaussian error envelopes with ranging from 2.5 mm to 20 mm. For different radii for the shells and spheres, r and respectively, (ranging from 35 mm to 90 mm), each m (ranging from 4 to 35 point pairs) simulated for a radius, 100 simulations, (where for ), are produced and registered for each . More briefly, the simulation encompasses iterations which are five loops deep:

  1. FOR a shell and sphere configuration.
  2. FOR different radii of the above configurations, r.
  3. FOR different numbers of homologous point pairs, m, for each r and configuration.
  4. FOR different homology errors envelopes for each m, r, and configuration.
  5. DO the creation of the data set to be registered and register them N = 100 times.
  6. Upon termination of each DO loop, calculate the mean, standard deviation, and statistic of the translation and rotation errors over the sample population of N = 100.

Six test points, two per axis, were symmetrically situated about the center of either the sphere or shell to encode the correct registration in the simulation for each A. No homology error was introduced to the test points so that the test points in each were exactly homologous to their pair in A. The correct translations, , and rotations, , about the same center of rotation were calculated for each using the test points. The simulated registrations of the N 's to the same A and their associated shifts, , and rotations, , were calculated and the uncertainty of the distributions (from their differences from and ) were found to ultimately yield and through each dimension and about each axis for each combination of factors. These simulations were written as batch files for Matlab and run on an SGI (Silicon Graphics Interface) workstation with twin 200 MHz Challenge processors.



Next: 5.3.2 Real Scans Up: 5.2 Methods Previous: 5.2 Methods


lukban@pet.mni.mcgill.ca
Wed Jan 18 14:28:16 EST 1995