This paper builds on the method of principal components analysis and its use for obtaining from a set of training image vectors a basis in which the members are rank ordered in terms of importance. The particular focus of this research is situations where the training vectors arise from images acquired at one range, call it the base range, and the image under question has been acquired at a different range. A natural question is whether one must train for all possible ranges. It is shown that, under certain assumptions, the eigenvectors for the data corresponding to ranges other than the base range may be approximated by performing a simple transformation on the eigenvectors derived from the training set at the base range. This is an important result, tending to obviate the need for acquisition of additional training patterns or for additional complex computations. It is also shown that under these assumptions the eigenvalues remain approximately constant over the different ranges even though the pixel size is changed. Bounds for the errors of approximation introduced by the method are derived. Numerical examples using image data are presented to demonstrate the application of the method. For this purpose a set of images was obtained via a detailed simulation of IR radiation from an M1 tank.
Principal components analysis: extension to scaling for application to images acquired at various ranges
1999
11 Seiten, 6 Quellen
Conference paper
English
Face Detection in Color Images Using Principal Components Analysis
British Library Conference Proceedings | 1999
|Face detection in color images using principal components analysis
IET Digital Library Archive | 1999
|Performance and ranges of application of various types of aircraft-propulsion system
Engineering Index Backfile | 1947
Principal Components Analysis Based Incident Detection
British Library Conference Proceedings | 1998
|