Schwarps: Locally Projective Image Warps Based on 2D Schwarzian Derivatives

摘要: Image warps -or just warps- capture the geometric deformation existing between two images of a deforming surface. The current approach to enforce warp’s smoothness is penalize its second order partial derivatives. Because this favors locally affine warps, fails local projective component image deformation. This may have negative impact on applications such as registration and deformable 3D reconstruction. We propose novel penalty designed smooth warp while capturing deformation’s structure. Our based equivalents Schwarzian derivatives, which are differential invariants exactly preserved by homographies. methodology derive set Partial Differential Equations with homographies solutions. call system equations we explicitly them for 2D functions using properties name Schwarp estimated penalizing residual equations. Experimental evaluation shows that Schwarps outperform in modeling extrapolation power, lead far better results Shape-from-Template camera calibration from