Quantifying coregistration errors in spectral imaging

It is well known that in multi- and hyperspectral imaging, the spatial coregistration of bands is important for the quality of image data. In practice, all spectral imaging sensors exhibit some degree of coregistration error, so that in any given pixel, different bands have a somewhat different "footprint" in the scene. Such coregistration errors may be in the form of differences in the position, size and shape of the spatial responsivity distribution, here termed pixel response function (PRF). There appears to be no standardized way to quantify the combined effect of these errors. This paper proposes a common metric for different types of coregistration error. Basically, the metric is the integrated difference between the PRFs of two bands in a given pixel. It is shown that under reasonable assumptions, this metric reflects the worst-case error in the signal resulting from coregistration errors between the two bands. To specify the coregistration of multiple bands, or in multiple pixels, an aggregate metric can be defined. The metric may be used for design optimization and should also be experimentally measurable. Extension to spectral and temporal coregistration is briefly discussed. The metric is proposed as a standardized way to report the coregistration performance of spectral image sensors.