Skip to content

← ArchivePaper2019

Strand-Accurate Multi-View Hair Capture

Giljoo Nam, Chenglei Wu, Min H. Kim, Yaser Sheikh

CVPRAcademic4 descendantsCFX

First method to capture hair with sub-millimeter strand-level accuracy from a multi-view rig using slanted stereo correspondences and mean-shift strand growth.

Abstract

This paper presents a method to capture high-fidelity hair geometry with strand-level accuracy from multi-view images. In the first stage, a line-based PatchMatch multi-view stereo reformulates traditional MVS with a slanted strand-line assumption, using a cost function combining photo-consistency and a geometric term that reconstructs each hair pixel as a 3D line and merges the depth maps into a point cloud with per-point line directions. A mean-shift based strand reconstruction algorithm then converts the noisy point data into a set of strands, and a multi-view hair growing step elongates short strands and recovers missing ones. Evaluated on synthetic and real captured data, the method reconstructs hair strands with sub-millimeter accuracy and pixel-accurate projection to novel views.

How to read this

Category
Capture method: multi-view hair geometry reconstruction
Contributions
  • A line-based PatchMatch multi-view stereo that reconstructs each hair pixel as a 3D line under a slanted strand-line assumption
  • A mean-shift strand reconstruction that converts noisy oriented point data into discrete strands
  • A multi-view hair-growing step that elongates short strands and recovers missing ones, reported at sub-millimeter strand accuracy
Context
Builds on structure-aware hair capture (referenced Structure-Aware Hair Capture, Luo et al. 2013), reformulating traditional MVS with a strand-line model for finer geometry.Builds on: Structure-Aware Hair Capture
Correctness
Evaluated on synthetic and real captured data with sub-millimeter accuracy and pixel-accurate reprojection; relies on a multi-view rig and the slanted-line strand assumption, so results depend on capture setup and may degrade on very occluded or fine wispy regions not stressed in the stated evaluation.
Clarity
Clear staged pipeline; a first pass conveys the line-MVS plus growth idea, a second pass is needed for the cost function and mean-shift formulation.
How to read it
Read the three-stage pipeline structure first, then a second pass on the photo-consistency-plus-geometric cost and the strand-growing step if you work on capture; check the evaluation conditions before trusting the accuracy claim for your setup.

Builds on

Built upon by

Related work

Keywords

This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →