Abstract
This paper describes a new approach to registration based on the theory of possibilities. We have focused on keypoints extraction methods which present the most important phase for registration which are: Surf, Corner and BRISK. The appearance of a large number of false bones has encouraged us to present a new approach that aims to reduce as much as possible the outliers. The class-based image registration based on the theory of possibilities increases the performance so it reduces the number of false keypoints found.