Quantification and reduction of erroneous differences between images in remote sensing

The reliability of image data, along with the difficulty of accurately comparing images acquired at different times or from different sensors, is a generic problem in remote sensing. Measurement repeatability errors occur frequently and can substantially reduce the system's ability to reliably quantify real spectral and spatial change in a target. This paper outlines methodologies for quantifying and reducing erroneous differences between monochrome and multispectral or hyperspectral images. [Each of these image types is acquired by an instrument that collects light photons across a variable range of the electromagnetic spectrum (often referred to as an image band). The hyperspectral image is often referred to as a hyperspectral cube with XY spatial dimensions and many Z bands (spectral demotions)]. In this paper, we specifically discuss the Pixel Block Transform (PBT), the Spectral Averaging Transform (SAT), and the Wavelet Transform (WAVEL). We briefly address sensor fusion. Results indicate that the PBT is a powerful cross-noise and repeatability error reducing tool, applicable to monochrome, multispectral, and hyperspectral images. The SAT is as powerful as the PBT in reducing error but is suitable only for hyperspectral imagery. WAVEL can reduce some of the finer scale noise, but it is not as powerful in reducing cross-noise as PBT or SAT and it requires some trial and error for selecting the appropriate wavelet function. It is important that those involved in developing statistically sound relationships between remotely sensed imagery and other data sources understand the problems and their solutions to prevent wasting time or developing relationships that are statistically insignificant or unstable, or that lead to faulty conclusions.