For image regression, a relation is established through regression between two images of different dates. The relation is used to predict pixel values in the second image, which are then subtracted from the first image. Regression reduces the effects of sen-sor, atmospheric and environmental differences between the two images. Development of suitable regression functions can be difficult.