Joint labelling and segmentation for 3D scanned human body

Hanqing Wang, Changyang Li, Zikai Gao, Wei Liang

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Citation (Scopus)

Abstract

In this paper, we present an approach to perform 3D human body labelling and segmentation jointly. Given a 3D mesh of scanned human body with texture, our approach segments it into 5 parts: head, torso, arms, legs and feet automatically. We assume that the faces on the same part of human body share similar color features and are constrained by geometry. According to this assumption, we formulate the labelling and segmentation of 3D Mesh as an energy function optimization problem. In this energy function, a data term models the color information and a smooth term models the geometry constraint. Then a GraphCut algorithm is applied to solve the optimization problem. The experiment results show good performance of our method.

Original languageEnglish
Title of host publicationSA 2016 - SIGGRAPH ASIA 2016 Virtual Reality Meets Physical Reality
Subtitle of host publicationModelling and Simulating Virtual Humans and Environments
PublisherAssociation for Computing Machinery, Inc
ISBN (Electronic)9781450345484
DOIs
Publication statusPublished - 28 Nov 2016
Event2016 SIGGRAPH ASIA Virtual Reality Meets Physical Reality: Modelling and Simulating Virtual Humans and Environments, SA 2016 - Macau, China
Duration: 5 Dec 20168 Dec 2016

Publication series

NameSA 2016 - SIGGRAPH ASIA 2016 Virtual Reality Meets Physical Reality: Modelling and Simulating Virtual Humans and Environments

Conference

Conference2016 SIGGRAPH ASIA Virtual Reality Meets Physical Reality: Modelling and Simulating Virtual Humans and Environments, SA 2016
Country/TerritoryChina
CityMacau
Period5/12/168/12/16

Keywords

  • 3D labelling
  • 3D mesh segmentation
  • GraphCut
  • Human body segmentation

Fingerprint

Dive into the research topics of 'Joint labelling and segmentation for 3D scanned human body'. Together they form a unique fingerprint.

Cite this