Change detection in multispectral images based on fusion of change vector analysis in posterior probability space and posterior probability space angle mapper

Abstract

Change vector analysis in posterior probability space (CVAPS) has been introduced recently as an effective method for change detection. CVAPS is based on the length of change and its direction in a posterior probability (PP) space. However, CVAPS is prone to similar direction cosine values. An approach to analyzing change by combining CVAPS and a new method called posterior probability space angle mapper (PSAM) is proposed in this study. PSAM establishes the similarity between two PP vectors of a pixel for two different dates by calculating the angle between them. This research presents a new change-detection algorithm based on combining CVAPS and PSAM (CVAPSAM), which is able to fully exploit change vectors in a PP space. While CVAPS uses a suitable threshold value to detect changes, CVAPSAM does not need to set a threshold. In addition, it reduces the similar direction cosine values source of error in identifying ‘from-to’ classes.