The Reduced Order Models (ROM) have been introduced as a low rank approximation of a studied dynamics to speed up its evalutaion. This low rank approximation is crucial for application related to CFD as the number of degrees of freedom limits their usage in design optimization or real time evaluation for dynamics and control. Even though some ROMs have been proven efficient on certain applications, their construction around linear latent subspaces reduces their usage to some specific industrial cases. In particular, it has been proven inefficient when dealing with convected dominated flows as they cannot be accurately described by a linear subspace. To overcome this issue we propose a general interpolation technique between snapshots of parametric simulation results. The core of this method relies on the computation and the use of spatial mappings to transport the snapshots over the manifold of solutions. This kind of approach is referred to as image registration and has been largely used in medical imaging. Today we propose to apply image registration to CFD data and compute a cheap interpolation in a non intrusive manner. The method starts with the use of markers to identify the coherent structures of the flow. From these markers, we will define the registration problem as the minimisation of an energy functional depending on the mapping we are looking for. The functional is designed to fill three objectives: align the markers, produce a bijective mapping through the regularization of its displacement field and impose on the interpolation trajectories to be compliant with the boundary conditions of the CFD domain. Finally, the solution should be computed fast enough so the interpolation method remains relevant for real time applications. For this, we constructed an appropriate compact research space for the velocity of the displacement field. This research space was built as the span of the first eigenfunctions of the Navier- Lam´e operator along with the appropriate boundary conditions. The regularization of the field was ensured by the minimization of its H2 norm. Finally, by applying this pseudo temporal mapping on the solution it is possible to construct non linear latent space or use it directly to perform a convex displacement interpolation of the snapshots.