1 June 2003 Multispectral illumination invariant classification with three visible bands and a near-infrared band
Author Affiliations +
Abstract
We describe multispectral image processing with three visible bands and a near-infrared band, whose pictures are taken by the multispectral sensor. We have improved multispectral classification for remote sensing to detect the target. The improved method features the division of the measured image by the target signature to remove the effect induced by illuminance change. The multispectral vector is transformed to angular space, so that the target position is always in (π/4, π/4) in angular coordinates. Then the translation of the angular coordinate is carried out so that the target position comes to the origin. The target class is determined by setting the decision boundary around the origin. The experiment reveals the effectiveness of this method.
© (2003) Society of Photo-Optical Instrumentation Engineers (SPIE)
Toshihiro Okamura, Toshihiro Okamura, Jun-ichi Kudo, Jun-ichi Kudo, Kunihiro Tanikawa, Kunihiro Tanikawa, } "Multispectral illumination invariant classification with three visible bands and a near-infrared band," Optical Engineering 42(6), (1 June 2003). https://doi.org/10.1117/1.1569497 . Submission:
JOURNAL ARTICLE
8 PAGES


SHARE
Back to Top