Radiometric Compensation in a Projector-Camera System Based on the Properties of the Human Visual System Dong WANG, Imari SATO,

Takahiro OKABE, and Yoichi SATO

June 25, 2005

Introduction

desired image directly projected

our system

our proposed method

camera

textured screen

projector

Our goal: Project onto textured

surfaces while preserving good photometric quality of input images

Previous Work [Nayar03] Project onto the textured surface Make the output captured by the camera

match the desired image

systemcompensated image

desired image

textured screen

Project

Capture

Match??

modified image

Modify

Project

directly

uncompensated

Limitation Physically limited dynamic range of the projector ⇒ Annoying artifacts

cutoff

modified imagecompensated image

The output of the projector saturates

Key Idea

Optimization problem: Minimize annoying artifacts by contrast reduction Maximize the contrast

⇒ Consider the properties of Human Visual System

poor contrast

cutoff remains

Visual Sensitivity [Bolin & Meyer 1998]]

Human Visual System is not sensitive to: High background illumination levels High spatial frequencies High contrast levels

original uniform sinusoidal straps noised

sensitive

not sensitive

How to Use the Loss of Visual Sensitivity?

Less contrast reduction effort where humans are less sensitive

⇒ How to determine the maximum error that can be tolerated?

Input

Maximum error

that can be tolerated

A Perceptually-Based Physical Error Metric

Predicts maximum luminance error that can be tolerated Makes use of threshold sensitivity, contrast

sensitivity, and contrast masking

LUMINANCE-DEPENDENT PROCESSING

SPATIALLY-DEPENDENT PROCESSING

TVI

CSF Masking

threshold map from TVI

elevation factor map

input image threshold map

[Ramasubramanian99]

Assumptions

Gray-scale images

Planar surface⇒ 　 homography for the geometric mapping of the pr

ojector-camera system

Lambertian surface

No ambient illumination

camera

textured screen

projector

Our Proposed Method

yx,

21 )],([ EyxEE

Error caused by artifacts Error caused by the degradation of the contrast

Total error

⇒ Determine the optimal global scalar α by minimizing the total error

α: global scalar for contrast reduction

λ: constant parameter

Results

desired image textured screen

compensated output image without contrast compression

uncompensated

Results

desired image

compensated output image without contrast compression

compensated output image with contrast compression α = 0.5678

threshold map

Conclusions Contributions

Incorporate the properties of Human Visual System into radiometric compensation

Relax a severe limitation of the radiometric compensation system

Physically limited dynamic range of the projector

Future Work Compensate color images

Use spatially varying scalars

Spatial temporal

Thank you