Bayesian Scale Space Analysis of Differences in Images

We consider the detection of image features that appear in different spatial scales or resolutions. In particular, the goal is to capture the scale dependent differences in a pair of noisy images of the same scene taken at two different instants of time. A new approach is proposed that uses Bayesian statistical modeling and simulation based inference. The method can be viewed as a further development of SiZer technology, originally designed for nonparametric curve fitting. A strength of the Bayesian simulation based approach is straightforward inference and modeling flexibility that facilitates the incorporation of domain specific prior information on the images under consideration. The method is believed to have applications for instance in satellite based remote sensing. In addition to artificial test images, a preliminary analysis of a pair of Landsat images used in satellite based forest inventory is therefore also included.