To perform a registration task, various types of diffeomorphic deformations can be combined to leverage their intrinsic properties (e.g., scales, linearity) or to incorporate prior knowledge (e.g., modules). To have a better interpretability of the model and to prevent the overlapping of the two types of deformations, their action on a shape have to be as “orthogonal” as possible. In this talk, I will first introduce the LDDMM model and its application for matching tasks, then i will present a quantification of the correlation between two deformations as the difference of the speeds of a varifold associated to a shape in the direction of both vector fields representing the two deformations.