Abstract
Spatial misregistration of multispectral scanner data exists when two (or more) spectral band signals supposedly representing the same location are in fact data values generated from two (or more) overlapping or entirely different ground locations. Misregistration can be found to occur between scanner data channels due to inherent scanner system characteristics and between time periods of multitemporal data due to errors in spatial registration. A study was conducted at the Environmental Research Institute of Michigan to determine what effect these forms of spatial misregistration may have on the accuracy of recognition processing of agriculturally oriented scanner data. A technique was devised and used to measure misregistration along a scan line between channels of data. Misregistrations in excess of one pixel in magnitude were detected in the scanner system examined. This led to the implementation of both analytical and simulation techniques to determine what effect varying degrees of misregistration would have on recognition accuracy and subsequent proportion estimation. Analyses found misregistration to severely reduce the availability of field center pixels and subsequently to introduce significant errors in the classification accuracy and correct proportion estimation of a scene containing an inflated number of mixture pixels. The results of the analysis emphasized the need for the elimination of misregistration in multispectral scanner data and a need for an awareness of the deleterious effects of misregistration in processing results.
Date of this Version
1976