Realigning 2D and 3D Object Fragments without Correspondences.

Area of Science:

• Computer Vision
• Geometric Modeling
• Image Analysis

Background:

• Reconstructing objects from fragments is challenging due to complex deformations.
• Existing methods often rely on point correspondences, which can be error-prone or unavailable.

Purpose of the Study:

• To develop a novel framework for simultaneous estimation of linear deformations and global non-linear transformations between object fragments.
• To enable accurate object reconstruction without relying on explicit point correspondences.

Main Methods:

• A general framework is proposed based on solving a polynomial system of equations.
• The method directly estimates alignment parameters, bypassing the need for feature matching.
• Quantitative evaluation on large synthetic 2D and 3D datasets, including rigid-body and affine transformations.

Main Results:

• The proposed algorithm accurately estimates deformation parameters for object alignment.
• Robustness against segmentation errors and numerical stability were exhaustively analyzed.
• Successful experiments were conducted on both 2D real images and volumetric medical data.

Conclusions:

• The developed method offers a robust and direct approach to estimating transformations for fragment-based object reconstruction.
• The framework demonstrates effectiveness across various dimensions and image types, including medical imaging.