Microscale Optical Capture System for Digital Fabric Recreation

Raúl Alcain

, Carlos Heras

, Iñigo Salinas

, Jorge López

and Carlos Aliaga

Departamento de Ingeniería Eléctrónica y Comunicaciones, EINA, Universidad de Zaragoza,

C/María de Luna 1, 50018 Zaragoza, Spain

Desilico Labs, C/Tellez 24, 28007 Madrid, Spain

Keywords: Optical Capture, Fabric Recreation, Cloth Rendering, Divergence Beam.

Abstract: Synthetic images are ubiquitous in the world, being extensively used in advertising, industrial design and

prototyping. However, automatic digital reproduction of fabrics is still an open problem in industrial contexts,

due to the inherent complexity of cloth appearance. We present a system to capture images of fabrics at micron

resolution, lit from a set of collimated beams of LED luminaires distributed along the hemisphere. The system

ensures plane wavefront with a 1:1 magnification while minimizing self-occlusions. It also allows for specular

and diffuse light components separation through polarization, has a very accurate focusing system and a

shallow depth of field for depth extraction. We demonstrate the system is suitable for later extraction of

geometric and optical properties of the cloth at the fiber level, which is the main requisite for high fidelity

photo-realistic cloth rendering.

1 INTRODUCTION

Cloth rendering is a very active research area in

computer graphics, and is becoming of increasing

interest of many other fields. This is because digitally

reproducing the appearance of fabrics has many

applications not only in the entertainment industry but

also in the context of textile design and

manufacturing. However, the appearance of cloth is

the result of complex light scattering interactions that

occur within the micro-structures present at the fiber

level in textiles (Aliaga el al. 2017). Thus, micro scale

optical capturing systems are needed for accurately

extracting the geometric and optical properties of the

fabric that allow to reproduce the appearance under

any lighting condition, allowing rendering solutions

that reach the scale of fibers, (hundreds of

micrometers), to gain accuracy and predictive power

(Zhao et al. 2011).

Most existing devices are targeted to capture

generic surface-like (without volumetric structure)

materials at a millimeter scale, usually focused in

extracting the surface normals and its reflectance, the

latest in the form of a Bidirectional Reflectance

Distribution Function (BRDF) that can be spatially

varying (SV-BRDF). For this, digital cameras are

used to capture images with basically two

approaches. One of them uses multiple viewing and

lighting directions covering the hemisphere centered

in the normal of the material surface in order to

recover the Bidirectional Reflectance Distribution

Functions (BTF). The other one relies on a single

fixed view and multiple light sources. All these

systems commonly work with small magnification

relations (Schwartz et al. 2014), since they are

interested in recovering the SV-BRDF or the BTF

because they need a number of different viewing

directions. This implies many issues related to depth

of field at grazing angles, particularly critical when

trying to reach very fine resolutions. Thus, these kind

of solutions are very useful for mid-distance (a meter)

material appearance modeling, but do not provide

small enough pixel size to take the real details, and

therefore are not well suited when the goal is to

extract geometric and optical properties for later use

in a realistic rendering context, for instance in the

case of realistic fabrics.

Very few works present a microscale optical

capturing system. A recent example is the work of

Nam and colleagues (Nam et al. 2016). They

implement cameras with macro lens, leading to very

short optical working distances that allow reducing

the dimensions of the system up to a few tens of

millimeters. Their system uses multiple LEDs and a

fixed viewing direction, working with up to 5:1

magnification. All components are built in a 40 mm

114

Alcain, R., Heras, C., Salinas, I., López, J. and Aliaga, C.

Microscale Optical Capture System for Digital Fabric Recreation.

DOI: 10.5220/0007356201140119

In Proceedings of the 7th International Conference on Photonics, Optics and Laser Technology (PHOTOPTICS 2019), pages 114-119

ISBN: 978-989-758-364-3

radius dome and captured sizes are of the order of 2

millimeters.

However, systems with such small dimensions

are not always suitable for predictive rendering in the

context of textile industry. The reason is that cloth can

be seen as a structure tiled along the material, and

such minimum tileable structure can often be much

larger than 1-2 mm for a great percentage of fabric

types. Thus, the capturing system needs a

compromise between magnification and sample size.

This is not simple to accomplish due to the thickness

that some fibers can present, some below 5 microns,

and the self-occlusions of the system at such short

working distances.

Relevant optical parameters for microscale

capturing systems are the divergence of the incident

lights and the acceptance angles for the captured light.

For industry purposes, exposure times become also

critical. All of them are related to the dimension of

the sample and the distances from the light sources

and the camera to the sample. As distances become

shorter and/or samples become larger, divergences

and acceptance angles increase. The impact of these

optical factors on the accuracy of the measured

reflectance can be relevant, impairing final results of

the rendering processes.

Having in mind these considerations, we present

in this paper a novel microscale optical capturing

system with low divergence (1º) of the incident light

and low acceptance angle (<8º) of the captured light.

It has 75 collimated high power white LEDs as light

sources at a distance of 650 mm from the sample. The

image is captured by a digital camera using 50 mm

focal objective with a 25 mm length extension tube.

This provides a working distance of 100 mm and at

the same time maintains x0.5 magnifications for

microscale captures, with an image size of 4x4

microns per pixel.

Because of the dimensions of the structure, the

optical system also makes possible a simple

polarization analysis by separating diffuse and

specular light components, and depth analysis using

focal stacks thanks to its very shallow depth of field.

The design of the system makes it robust to vibrations

and takes into account all the mentioned optical

factors in order to assure a suitable system specialized

for capturing fibrous materials like fabrics. It

improves previous approaches for such goals, and it

is optimal for capturing high quality data that can be

used to achieve realistic rendering fabrics.

2 MESUREMENT SYSTEM

The microscale optical capturing system consists of

four main different parts: mechanical structure,

lighting system, image capture system and holder.

2.1 Mechanical Structure

The whole system (Figure 1) is a 1.2 m diameter

hemisphere-shaped structure, built with black

anodized aluminum profiles. There are nine

aluminum arms, formed by nine sections, which

support the “theoretical” 81 light modules pointed

towards the target fabric. Due to the occlusion of the

camera, it will eventually be 75 light.

This design removes most secondary reflections,

which could interfere with a measurement, adding

wrong illumination to the fabric to be sampled.

Figure 1: Optical capture system.

This structure is completely isolated from the

fabric holder and the camera support, in order to

avoid any mechanical vibrations during the capture.

2.2 Lighting System

Each of the light modules consist of several pieces: a

high power white LED, a collimating lens, a linear

polarizer, and a 3D printed case (see Figure 2). These

cases are attached to the structure at regular intervals

and can be suitably aligned towards the target.

The lighting modules have been designed to

produce a low divergence incident beam. The 20 mm

effective diameter, 16 mm focal lenses, combined

with the 3x3 mm LEDs, result in an exit beam

divergence of 5º (blue in Figure 2):

𝐷 =

𝑆𝑜𝑢𝑟𝑐𝑒 𝑟𝑎𝑑𝑖𝑢𝑠

𝑓𝑜𝑐𝑎𝑙 𝑙𝑒𝑛𝑔𝑡ℎ

3𝑚𝑚

16𝑚𝑚

= 0.09 𝑟𝑎𝑑

≅ 5°

(1)

Microscale Optical Capture System for Digital Fabric Recreation

115

However, as the size of the captured target image

is 10x7.5 mm, at a distance of 650 mm from the

source, the divergence of the incident beam is 1.3º

horizontal and 1º vertical ( green in Figure 2):

𝐷

′

𝑆𝑎𝑚𝑝𝑙𝑒 + 𝑙𝑒𝑛𝑠 𝑑𝑖𝑎𝑚𝑒𝑡𝑒𝑟

2 𝑥 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒

10 + 25𝑚𝑚

1300𝑚𝑚

= 0.03 𝑟𝑎𝑑 ≅ 1.5°

(2)

In the end, the effective divergence on the target

will be lower, as this is only the divergence of the

limiting rays. In Figure 2, yellow lines corresponds to

a perfect collimated beam from a point light source.

Figure 2: Lighting modules and beam divergence.

The linear polarizer can be manually rotated to

adjust the polarization to that of the polarization

analyzer placed on the camera, allowing to separate

light components in later post-processing.

Additionally, the LEDs are uniformly distributed at

18º intervals, making possible the illumination of the

target with many incoming light directions over the

hemisphere.

2.3 Image Capture System

Given the thickness of the individual fibers forming

the cloth, most of the times in the order of microns,

we have designed the capture system with a

resolution of 4x4 microns per pixel. At the same time,

the system should be able to capture images of around

10x10 mm in size, in order to determine the fabric

structure of the yarn level, which is usually in such

range of sizes.

We use a Mightex SME-B050 monochrome

camera, with a 2560x1920 resolution and a 2.2x2.2

um pixel size. A 50 mm focal lens with a 25 mm

extension space provide 0.5 magnification at a

distance of 100 mm. The space works as an extension

tube and is key to achieve a large working distance

without losing magnification. Such large distance is

required to avoid the systems to cast shadows on the

Figure 3: Image capture system.

fabric sample, that is, reducing light occlusions to the

minimum possible.

A multi-position motorized filter slider with three

RGB filters has been placed at the space between

camera and lens. With these filters, we can perform

color analysis without the loss of resolution due the

placement of the pixels in the mosaic of a color sensor

(Bayer pattern).

In addition, a polarization analyzer consisting of

two crossed linear polarizes can be placed in front of

the lens, with the aim of separating specular and

diffuse light component, taking advantage of the

depolarizing nature of light paths after a number of

bounces or interactions with the medium.

Finally, the whole capture system is mounted over

a motorized linear travel stage, providing automatic

camera focus. This also allows to perform depth

analysis of the fabric samples by using focal stacks

and depth from defocus techniques.

2.4 Fabric Holder

The fabric holder (Figure 4) can be displaced using an

XY translation mount to capture different fabric

sections. It can also be rotated to capture both sides of

the sample. The holder has also been design to meet

the requirement of minimal light occlusions that

could shadow the target.

3 CALIBRATION AND QUALITY

TESTS

This section describes the series of tests performed to

ensure the correct performance of our measurement

system.

PHOTOPTICS 2019 - 7th International Conference on Photonics, Optics and Laser Technology

116

Figure 4: Fabric holder.

3.1 USAF Resolution Test

The optical resolution of the complete optical system

has been measured, using the standard 1951 USAF

resolution test (Figure 5 a), to be 7-5, which is

equivalent to 203 line-pair/mm. Therefore the

resolution of the system is limited only by the camera

pixel size (454 pixels / mm), and not by the lens and

extension used.

Figure 5: Captures of 1951 USAF test (a) and spacing grid

displaced (b), and corrected (c).

3.2 Fixed Frequency Grid Distortion

Targets

A 125 µm uniform spacing grid has been used to

evaluate image distortion and chromatic aberration

and displacement. This test shows that the distortion

is under the resolution level of the system in the whole

image.

We observe, however, chromatic aberration and

image displacement when using different RGB filters.

Both are repetitive and can be easily corrected (Figure

5 b,c). The chromatic aberration is corrected by

changing the working distance of the camera for each

filter with the motorized linear stage. The

displacement is fixed during image processing.

These corrections remain constant, so they only

had to be determined once, during a calibration

process.

3.3 LED Uniformity Analysis

The lighting system uses XP-L2 Cree white LEDs.

We have tested their uniformity, both in power and

color, using a Laser 2000 Smini spectrometer, with a

225 to 1000 nm spectral range. The average color

temperature of the LED batch was 5639 K with a

standard deviation of 367K (Figure 6). The luminous

flux was 245 lm on average with a standard deviation

of 43 lm.

These variations have been calibrated and the

results used during the image processing stage to

correct non-uniformities.

Figure 6: LED color temperature calibration.

4 CAPTURED IMAGES

We show below some of the images captured with our

systems. Figure 7 shows images obtained directly

from the camera for a given illumination angle.

Already showing how some specific types of fibers

(e.g. fly out fibers) emerge under particular lighting

conditions.

For geometric extraction purposes, diffuse

lighting becomes very useful instead of direct

collimated light. Taking advantages of the additive

nature light, Figure 8 show the composite images

obtained combining the frames taken under each of

the 76 possible incoming light directions.

(a)

(c)

(b)

Microscale Optical Capture System for Digital Fabric Recreation

117

Figure 7: Images of a sample under four different incoming

light direction.

Figure 8: Composite images of the 75 different angle

captures. The black images of the center are due to the

occluded LEDs.

5 RENDERING RESULTS

The data obtained with the presented optical capture

system are well suited to extract the geometric and

optical properties of the fabric at the fiber level. These

properties are then used as input for a photo-realistic

rendering engine based volumetric path tracing, and

the preliminary tests show compelling results (Figure

9). Essentially, the engine simulates the light

transport at the scale of fibers, also modeling the

anisotropic light scattering patterns at micron scale.

Such properties are properly extracted with the

presented device, which provides enough resolution,

small enough pixel size and good level of light

collimation to meet our requirements.

Figure 9: Synthetic recreation of a garment using our

rendering engine and the geometric and optical parameters

extracted by the presented optical capture system.

6 CONCLUSIONS

We have designed and built a microscale optical

capture system. It meets the resolution and size

requirements for a realistic cloth rendering, while

keeping the incident beam divergence around 1º.

The design of the system makes it robust to

vibrations, and a proper calibration of the system

corrects all deviations due to LED non-uniformities,

and lens and filters chromatic aberrations and

displacements.

The system makes possible a polarization analysis

of light for specular and diffuse reflection

discrimination, also allowing to perform a depth

analysis using focal stacks of the fabric thanks to a

precise control over the working distance of the

camera. The captured images demonstrate the system

reaches the required quality for fabric digital

reproduction purposes, as the preliminary renderings

using this technology show.

REFERENCES

Aliaga, C., Castillo, C., Gutierrez, D., Otaduy, M. A.,

Lopez-Moreno, J., & Jarabo, A. (2017). An Appearance

Model for Textile Fibers. Computer Graphics Forum.

https://doi.org/10.1111/cgf.13222

Nam, G., Lee, J. H., Wu, H., Gutierrez, D., & Kim, M. H.

(2016). Simultaneous acquisition of microscale

reflectance and normals. ACM Transactions on

Graphics. https://doi.org/10.1145/2980179.2980220

Schwartz, C., Sarlette, R., Weinmann, M., Rump, M., &

Klein, R. (2014). Design and implementation of

practical bidirectional texture function measurement

devices focusing on the developments at the University

of Bonn. Sensors (Switzerland).

https://doi.org/10.3390/s140507753

PHOTOPTICS 2019 - 7th International Conference on Photonics, Optics and Laser Technology

118

Zhao, S., Jakob, W., Marschner, S., & Bala, K. (2011).

Building volumetric appearance models of fabric using

micro CT imaging. ACM Transactions on Graphics,

30(4), 1. https://doi.org/10.1145/2010324.1964939

Microscale Optical Capture System for Digital Fabric Recreation

119