A Hybrid Diffusion Model for 2d Dense Motion Estimation

2D motion field is the velocity field which presents the apparent motion from one image to another in an image sequence. In this paper, with the objective of accurate estimation of 2D dense motion field, a hybrid diffusion model is proposed. The present approach differs from those in the literature in that the diffusion model and its associated objective functional are driven by both the flow field and image, through the nonlinear isotropic diffusion term and the linear anisotropic diffusion term, respectively. The diffusion function in the model is required to be non increasing, non negative, differentiable and bounded. Furthermore, using Schauder’s fixed point theorem, we prove the existence, stability and uniqueness of the solution to the proposed hybrid diffusion model. The semi-implicit scheme is proposed to implement the hybrid diffusion model. We demonstrate its efficiency and accuracy by experiments on synthetic and real image sequences.