Image morphing using deformation and patch-based synthesis

Image morphing creates a smooth transformation between images. Traditional morphing techniques may fail when the two input images have large displacements. In this paper, a two-step scheme is proposed to deal with this challenging task. Firstly, a series of deformed images is generated through optimizing an adaptation of deformation objective that composed of as-rigid-as-possible energy, group lasso penalty, and smoothness penalty. Then, the forward and backward deformed sequences are aligned utilizing the patch-based synthesis, satisfying source similarity and temporal coherence. Guided by a few user-specified correspondences, the proposed method works well in many morphing scenes.