Thin plate splines, image registration and scientific visualization
Date
1992
Authors
Berkley, James Michael
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Thin plate splines are excellent tools for image registration, where exact interpolation is desired, but where there are local distortions in the data. Thin plate spline matrices are extremely ill-conditioned, but we have found that we can improve the condition number significantly through scaling onto the unit square and through ensuring a minimum separation between control points.
As with all interpolating functions, thin plate splines are subject to overshoot difficulties. In this thesis, we have investigated some techniques using scientific visualization for detecting severe interpolation overshoot.
Two practical applications of thin plate spline registration are examined, Change Detection in Side-Scan Sonar Images and Survey Monument Registration. Thin plate spline warping is very effective at registering two side-scan sonar images for comparison, in spite of severe local distortions due to sensor movement. Because of limitations in the data available, thin plate splines are not as effective at transforming between the North American Datums of 1927 and 1983 for survey monument registration.
Description
Keywords
UN SDG 14: Life Below Water