Abstract

We present a novel fully-automated approach to non-rigid registration for high-resolution facial scans using conformal harmonic maps. The novelty of this paper is its use of applied deep learning models to prepare data for geometric algorithms to compute non-rigid registration. We use facial detection to both constrain the boundary of the face and provide a mechanism to manipulate the input mesh. We use conformal harmonic maps[7] to map a dense 3D point cloud to the closed unit disc D1(0) and optimize the weights of each edge. Our experiments show the effectiveness of this approach.

Year

5-2023

Document Type

Thesis

Keywords

Registration, Non-Rigid Registration, Deep Learning, Conformal Harmonic Map

Degree Name

Master of Science (MS)

Department

Computer Science

Advisor

Xianfeng Gu

Share

COinS