[其它] Towards Hairstyle Reconstruction Using Thermal Imaging

[复制链接]

彬彬

797 主题	1 听众	1万积分

资深设计师

Rank: 7 Rank: 7 Rank: 7

纳金币: 5568
精华: 0

电梯直达

楼主

发表于 2012-1-4 17:00:35 |只看该作者 |倒序浏览

Abstract

The reconstruction of hairstyles is one of the most challenging tasks

in the field of hair modeling.

We present a novel technique for generating fiber-based hair models

from video sequences captured by a hand-held infrared camera.

The major advantage of using thermal imaging lies on the fact that

hair is actively re-emitting heat from the scalp and thus several issues

of conventional setups illuminating the hairstyle from outside,

such as shadowing and anisotropy in reflectance, can be avoided.

First promising results indicate the potential of our method.

Keywords: hair simulation, image-based modeling, 3-D reconstruction,

infrared imaging

1 Introduction

An important and very challenging task in the domain of hair modeling

is the creation of hairstyles. To date, this process commonly

involves lots of manual work by skilled 3D artists.

Recently, several techniques have been presented that aim to model

hairstyles from images, either by attempting to capture shape

directly [Paris et al. 2008; Wei et al. 2005] or by employing

physically-based models for user guided synthesis [Wither et al.

2007]. However, all these techniques suffer from shortcomings regarding

automated hair synthesis that will be discussed in the following.

While forWither et al.[2007] user interaction is part of the concept,

the two image based approaches [Paris et al. 2008; Wei et al. 2005]

are prone to shadowing and segmentation issues. Moreover, in particular

in case of light colored hair the two methods are likely to fail

as orientation fields, that are used to drive the hair synthesis, cannot

be faithfully estimated due to strong multiple scattering effects

inside the hair volume. Moreover, for Paris et al. [2008] a dedicated

setup including multiple registered cameras, projectors and

light sources is required. Even though their results are quite impressive

these hardware requirements as well as the computational

effort reduce practicability of the method.

The results presented by Wei et al.[2005], where video recordings

of a single hand-held camera are used as input, show topological

errors. Moreover, all required segmentation, which is far from being

trivial using an automated method, was done manually. Since

a hairstyle is modeled based on its visual hull, the method fails to

capture more complex features such as concavities and curls.

Using thermal image data as input several issues can be avoided as

hair is actively emitting radiation. This gives major advantages over

previous image-based techniques:

With the hair as “light source” shadowing issues can be completely

avoided.

We do not have to deal with complicated hair reflectance (i.e.

strong directional highlights and multiple scattering): Heat is

emitted from the head, absorbed by the hair and causes nearly

isotropic emission.

The segmentation of the hair region can be based on the temperature

values and is hence very simple and robust.

The temperature distribution over the hairstyle gives valuable

information about the distance of hair fibers and scalp.

2 Background

Every object with temperature above the absolute zero emits electromagnetic

radiation, and the amount of irradiated energy is

directly proportional to its temperature [Dereniak and Boreman

1996]. A modern thermographic camera captures radiation in the

mid-infrared range of the electromagnetic spectrum (900–14000

nm), detecting temperatures between 240 K and 670 K with a resolution

of 0.1 K This interval more than covers the potential range

of temperatures occurring for hair.