FreeYourEyes:RetinalImageDeblurringDisplaywithEnlargedViewingZone

XuanerCeciliaZhang*,Jinkyu Kim*,LauraWaller,BrianA.Barsky,RenNg

DepartmentofElectricalEngineeringandComputerScience

UCBerkeley,CA94720

*Theseauthorscontributeequallytotheproject

Light field vision-correcting displays improve visual acuity by generating a

corrective 4D light field, which produces a deblurred retinal image without the

need of physical optical corrections such as eyeglasses and contact lenses.

Introduction

Runtimecomparison

#ofpixelsonadisplay

#ofpixelsofapinhole

102k

5x5

409k

10x10

1,638k

20x20

6,553k

40x40

Huangelal

Proposed

386.9

0.61

541.6

2.01

763.4

8.11

860

33.49

Discussion

SystemDesign

In our system, free eye movement is limited by the specified viewing zone. This viewing zone could be potentially extended using an eye-

tracking system. Along with using a pinhole mask for a parallax barrier based vision-correcting display, it would be worth combining the

display with lens-let arrays for better brightness or a multi-directional diffractive backlight technology that supports high angular resolution.

LightFieldRenderingAlgorithm Analysis

The central concept of a corrective light field rendering is

to compute each pixel value on display by sampling the

light rays travelling through each pixel.

References[1] V. F. Pamplona, M. M. Oliveira, D. G. Aliaga, and R. Raskar. Tailored displays to compensate for visual aberrations. ACM Trans. Graph., 31(4):81, 2012

[2] F.-C. Huang, G. Wetzstein, B. A. Barsky, and R. Raskar. Eye-glasses-free display: towards correcting visual aberrations with computational light field displays. ACM Trans. Graph.,

33(4):59, 2014

The Key Idea: Rendering a 4D light field of a virtual scene placed at the focal

plane so that a viewer can perceive the image in focus without the need of optical

correction.

Algorithm 1: 1D ray tracing for light field rendering with local viewer

1 Set a viewing zone and find its center position Ox;2 Connect each pinhole tj with viewing center Ox to find the correct sets ofLCD pixels vj−k:j+k for pinhole tj

3 Sample N=25 points vnm, n = 1, 2, . . . N for each pixel vm;4 Trace a ray passing pixel tj and sample vnm to intersect with focal planeat pixel pj ;

5 if intersect with focal plane then

6 for R,G,B channels do

7 assign pixel intensity I(pj) to sample vnm;

8 else

9 set sample vnm to black ;

10 Calculate pixel value vm asP

n vnm

Figure2. DesignoftheproposedvisioncorrectionsystemFigure1. Myopiaandhyperopiaillustration

Figure4.Illustrationoftwoapproaches

Virtual Image Generation:

Since a rendered light field is limited by the

bandwidth of the display, a part of the virtual

image may be blocked as the viewer moves,

we propose two approaches:

1) Content-preserving that rescales the virtual

image so that the viewer can see the entire

image within a range of viewing zone

2) Content-cropping that maximizes the use

of display bandwidth but causes content

loss

Problem: Previous works present a promising way to correct viewer's eye

aberration at a fixed viewpoint, leaving the challenge of a real-time system that

supports eye movement.

Solution: We propose a light field vision-correcting display that efficiently

renders a corrective 4D light field to compensate for low-order eye aberration

(nearsightedness and farsightedness) over a viewing zone, in which the eye is

free to move without losing visual acuity.

ResultsThe proposed method

• shows comparable visual performance at most of the viewpoint

• shows comparable visual performance at most the viewpoints in

terms of SSIM and PSNR

• take 33.49 seconds to process over 4 million light rays

(under MATLAB environment, with a 3.1GHz Intel Core i7 with 4GB RAM)

More intuitively, to compensate for a reasonable amount of eye aberration,

for a screen size of an iPhone 6s (4.7 inch diagonal) we require a ~3133 by

~1762 resolution (current iPhone 6s resolution is 1334 by 750).

Figure3.Wideandnarrowviewingzone

Spatial and angular resolution

requirement:

Given the viewer’s severity of

defocus blur and the viewing

distance, the depth of a virtual

image can be determined.