Non-rigid registration under anisotropic deformations

Roberto M Dyke,Yu-Kun Lai,Paul L Rosin,Gary K.L Tam

doi:10.1016/j.cagd.2019.04.014

Roberto M Dyke, Yu-Kun Lai + Show 2 more

Open Access

https://doi.org/10.1016/j.cagd.2019.04.014

Copy DOI

Abstract

Non-rigid registration of deformed 3D shapes is a challenging and fundamental task in geometric processing, which aims to non-rigidly deform a source shape into alignment with a target shape. Current state-of-the-art methods assume deformations to be near-isometric. This assumption does not reflect real-world conditions, for example in large-scale deformation, where moderate anisotropic deformations (e.g., stretches) are common. In this paper we propose two significant changes to a typical registration pipeline to address such challenging deformations. First, we introduce a method to estimate anisotropic non-isometric deformations and incorporate this into an iterative non-rigid registration pipeline. Second, we compute additional correspondences in non-isometrically deforming regions using reliable correspondences as landmarks and prune inconsistent correspondences. We compare the performance of our proposed algorithm to several state-of-the-art methods using existing benchmarks. Experimental results show that our method outperforms existing methods.

Highlights

Surface registration is a fundamental problem in the domains of computer graphics and vision, in which the aim is to find a transformation that best aligns two input surfaces
Because of the simple way correspondences are generated, N-Iterative Closest Point (ICP) is fast enough to be used in some real-time applications; though alone, it is incapable of coping with large-scale deformations
To cope with large deformations, we propose a new consistency measure that takes into account anisotropic non-isometric deformation explicitly

Summary

Introduction

Surface registration is a fundamental problem in the domains of computer graphics and vision, in which the aim is to find a transformation that best aligns two input surfaces. In many real-life scenarios, surfaces are often non-rigidly deformed. Non-rigid surface registration is required to find the non-rigid transformation between them. Extending from the well-known Iterative Closest Point (ICP) approach for rigid registration Besl and McKay (1992); Chen and Medioni (1992), Non-rigid ICP (N-ICP) methods Bouaziz and Pauly (2013) achieve registration for non-rigid surfaces by alternating between two steps. A set of correspondences is computed using a closest point criterion, and the second step identifies a non-rigid transformation that minimises an error metric. Because of the simple way correspondences are generated, N-ICP is fast enough to be used in some real-time applications; though alone, it is incapable of coping with large-scale deformations.

Methods

Results

Conclusion