An evolutionary algorithm for the registration of 3-d surface representations

Evolution strategies based image registration via feature matching

“…Fischer et al studied the 3D image registration problem as part of object recognition. An evolutionary algorithm-based 3D surface registration method was presented in Fischer et al's work [10].…”

“…The average number of generations needed to achieve convergence was in the hundreds, but in extreme cases [10], it could be thousands.…”

<title>Volumetric image registration by template matching</title>

“…Typical approaches are hill climbing, 28 gradient descent, 10 simulated annealing, 21 and evolutionary algorithms. 12 Such algorithms often require that a good initial registration be provided. They then determine globally optimal parameters to register the images.…”

“…A minimum is obtained when sum of absolute di erences or geometric distance is used as the similarity measure, while a maximum is obtained when cross-correlation coe cient or mutual information is used. A tri-quadric function is de ned by S(x y z) = Ax + Dxy+ E y z+ F x z + Gx + H y + I z+ J: (12) Parameters A{J are determined by the least squares method using the 3 3 3 array of similarities and their coordinates. Once A{J are determined, the maximum (minimum) of S(x y z) is determined by computing its derivatives with respect to x, y, a n d z setting them to zero and solving the obtained system of linear equations for x, y, a n d z.…”

Automatic Registration of Range Images by Using Shape-Index-Based Artificial Images

“…Various improvements and variants of the original ICP has been proposed, including the iterative closest compatible point algorithm (ICCP) [2] , which verifies the closest point pairs by additional attributes like color or surface normal. Furthermore, more sophisticated optimization schemes are proposed such as the simulated annealing [3] and the evolutionary algorithm [4] . Rusinkiewicz and Levoy [5] and Rodrigues et.…”