Radiometric normalization between multitemporal image sets is a useful pre-processing step prior to change detection analysis. Radiometric normalization between multitemporal image scenes containing areas of natural land cover and habitat is most effectively accomplished using a histogram matching approach that normalizes image scenes based upon values of global mean and standard deviation (not reshaping of histograms, as is common with histogram matching functions in many image processing software algorithms). The assumption for this approach is that changes within the scene are few and localized. This approach is efficient and follows a simple processing flow. Multitemporal image scenes are masked so that they cover the same geographic area and only habitat reserve lands comprise the scene. The global mean and standard deviation statistics are extracted and utilized in a single model which adjusts one image so that its mean and standard deviation values match those of the other (reference) image. The image with the highest standard deviation is chosen as the reference, so that there is no reduction of radiometric resolution and information content.

A potential disadvantage of the histogram matching approach is that large area changes between multitemporal image scenes may result in real differences in image mean and variance which should be maintained. Quality control by inspection of the image scenes is required to ensure that there are no significant differences (such as large area land cover change or cloud cover) affecting the statistics of the images being normalized. If large area changes are present, data from these areas may be excluded during the generation of statistics for the histogram matching process.

The images to the left represent radiometrically normalized 2001 ADAR 5500 image mosaics and 2000 radiometric reference mosaic. Pseudo-invariant feature histogram matching radiometric normalization techniques were compared. Display is false color infrared. Each mosaic is displayed using the same contrast stretch. To view a larger version of the images just click on the image to the left.