Change detection from remotely sensed multi-temporal images using morphological operators

This paper presents a new approach to spatial change detection. The algorithms developed are based on the use of basic morphological filters and on more advanced concepts such as geodesic transformations. Such techniques are able to overcome the traditional problems associated with change detection from remotely sensed multi-temporal images. As a matter of fact it is already known that traditional methods using the concept of direction variation of the change vector are inadequate for a precise detection. These frequential techniques lay on very limitative statistical hypotheses: gaussian distribution, a priori determined ratio of change, very large images and relatively small ratio of change. However in a prior study, it was determined that the basic operators of mathematical morphology were partly corrupting the results of change detection by introducing bias on the shape of the objects and also by shifting the edges in the displacement direction of the structuring element. The geodesic transformations are correcting these topological difficulties in an elegant way. The usual threshold step is replaced by appropriate structuring element interval of sizes. Thus, it becomes possible to treat the spatial change detection problem by using a single formalism. The first results show that the particles of change are detected even for very slight radiometric variations, with the advantage of taking into account the configuration of the neighbourhood.