3M’s Display Quality Score · 3M’s Display Quality Score 3M’s tool to guide the development...

3M’s Display Quality Score3M’s tool to guide the development of higher quality displays

How can we build a great display?

© 3M 2015. All rights reserved. 3M Confidential. 3

How can we build a great display?


Approach 1: Copy the physics

Limited by current technology and efficiency considerations


Approach 2: Recreate Visual Response

Because the visual system does not encode all of the information in the light field, recreating a visual experience does not require recreating the light field with perfect fidelity


Approach 2: Recreate Visual Response

DQS uses knowledge of the visual system and visual preferences to guide developers toward displays that create superior visual experience

Quantitative accurate prediction of how human quality judgment is influenced by engineered display parameters (resolution, luminance, etc.) and thereby avoid costly experimentation with human observers

Goal of the 3M DQS

7

Allow display engineers to identify display attributes and combinations of attributes that have the largest impact on image quality

Purpose of the 3M DQS

8

Background and Current Status

Image quality metrics for hardware and software identify 6 dimensions underlying image quality judgment1 : • Color, Luminance (Brightness/lightness), Contrast,

Resolution/Sharpness, Arifacts (e.g. non-uniformity, noise), Parameters of the display/print media (e.g. screen gloss)

DQS v1.0 characterizes quality from Color, Luminance, Contrast, Resolution and Display SizeArtifacts and media influences are being considered for DQS 2.0

9

How does DQS work?

How does DQS work?

DQS relies on well established facts about the visual processing and visual preferences

These facts are described in mathematical models

To understand how DQS works, we need to understand some basic facts about the visual system

Overview

Present and explain some of the key empirical bases for DQS, explain why they are important in display development and describe how they are modeled

Describe some of the data and modeling methods used in 3M’s development of DQS

Explain how to use and interpret DQSs

Human visual processingThe Basics


Visual experience is created by the pattern of activity in neural networks that start in the eye and project across many regions of the brain

Human visual processing1. Transducing light into a neural response


Fovea: central high resolution vision

Neural network in the eye

Human visual processing2. What detail can the visual system resolve? The Contrast Sensitivity Function


Increasing spatial frequency

Dec

reas

ing

cont

rast


More contrast is required for the visual system to detect fine details

Campbell & Robson, 1968


2.9 x 10-4

cd/m2

290 cd/m2.

van Nes, M. A. Bouman (1967)

Sensitivity to different levels of detail depends on luminance

1. Greater image detail can be seen at higher luminance

2. Small scale artifacts are less visible that larger scale


DQS employs a model of the CSF and its changes with luminance to predict visibility of patterns in displays

We can model this!

Human Visual ProcessingHow does the visual system create color from light?

Trichromatic Basis of Human Color Vision

Color is our experience of the relative response of the 3 cone types

Any light spectra creating the same relative response should appear the same

Summary of DQS Ingredients Forced Choice

Experiments

Data & Model on

human vision

Data & Model on

display system

Experimental MethodologyForced choice paradigms

Do you prefer the image on the left or the image on the right?

How Image Pairs Were Selected: Method 1Pre-select images and perform all pairwise comparisons

A B

C D

A B C D

A .5 .7 .1 .95

B .3 .5 .25 .55

C .9 .75 .5 .95

D .05

.45

.05

.5

Table cells show (fictional) proportion that stimulus in row was selected over stimulus in column

Especially useful when we do not know what direction people’s preferences will go (e.g. Will they prefer less or more luminance under dim illumination?)

“Unidimensionalscaling” used to recover preference scale

How Image Pairs Were Selected: Method 2

Vary one of two images along one dimension (e.g. color gamut or luminance) to determine the size of the difference along that dimension that gives some proportion of times the fixed “standard” is preferred to the varied image (“Staircase procedures”)

Useful when can assume preference “transitivity” along the dimension being varied (e.g. if a medium color gamut is preferred to small gamut and large is preferred to medium, then large will be preferred to small)

standard

comparisonProportion standard

preferred to comparison.55

.72

.96

Experimental Set-Up

High color gamut displays (HP Dreamcolors)

Viewing conditions: Indoor lighting

Data collected from 6 countries

28 images calibrated to meet display parameters


Proportion of people preferring larger over smaller color gamut. The 0 point on the x-axis represents the smallest gamut tested (~72% NTSC)

Results: All countries


Results by country

Examples of DQS Model Fits to Color Preferences Data

“Staircase data” Preselected image pair data

‘standard’ gamuts

Examples of DQS Model Fits to Color Preferences Data

“Staircase data” Preselected image pair data

14 participants (USA)>3,000 observations

212 participants6 countries, 8 locations>58,000 observations

Key points

1. Data directly show the proportion of people prefer different color gamut2. The vast majority of people consistently prefer large over small3. ~14% of the population is indifferent

Summary of Development of the DQS

Survey of existing metrics indicated Barten’s 19872 square root integral (SQRI) was best suited for characterizing display image quality• SQRI uses the human contrast sensitivity function to predict the effects of resolution, image

size, luminance and luminance contrast• Early DQS studies on luminance (viewing angle) and image size verified SQRI correlation with preference

Limitations of SQRI: Color is absent and we know color impacts perceived display quality

3M study and other studies indicate that effects of black point and contrast on preferences not adequately predicted by SQRI

3M improves on SQRI metric based on color and black point research

2P. Barten, “Evaluation of subjective image quality with the square-root integral method,” JOSA A, vol. 7, no. 10, pp. 2024–2031, 1990.

DQS: Current Form

Barten’s SQRI integrated influence of: Luminance, Size, Resolution, Contrast

Influence of color gamut and black point

Overview of DQSThe User’s perspective: Calculation and Interpretation

Diagonal screen size & viewing distance

# pixels (horizontal/vertical)

Luminance (black & white)

Chromaticity of 3 primaries

Display type (LCD, OLED)

DQS Input Parameters

Landmark Examples of Absolute DQS

Diagonal Dist pixels H pixels V white black color DQSExtreme TV 85 108 4096 2160 600 0.02 135.0% 50High end TV 65 108 4096 2160 500 0.17 87.2% 42.3

High Average TV 55 108 1920 1080 450 0.18 87.2% 37.2Low Average TV 32 108 1920 1080 300 0.38 78.5% 26.0Great notebook 15.6 22 4096 2160 350 0.35 120.0% 45.5

Low end notebook 15.6 14 1366 768 250 0.5 50.0% 21.5Great smart phone 5 14 1920 1080 550 0.55 120.0% 36.1Ave. smart phone 4.3 14 800 460 400 0.5 60.0% 20.1

Old cell Phone 1.6 14 320 240 250 2.5 25.0% -11.9

Interpreting the DQS

DQS provides an absolute score but, from an engineering, manufacturing and marketing perspective, the meaning is in the differencebetween scores

ΔDQS

Proportion of Population

Preferring One Over Other

-2 ≤14%

-1 ≤25%

0 50%

1 ≥75%

2 ≥86%

DQS as a tool to guide display development

DQS Predictions: Resolution

Device type LCD

Diagonal 48”

viewing distance 275cm

black point .5 cd/m2

white point250

cd/m2

color gamut sRGB

DQS Predictions: Luminance

Device type LCD

Diagonal 48”


# pixels horizontal 1920

# pixels vertical 1080


color gamut sRGB

DQS Predictions: Color

Device type LCD

Diagonal 48”





white point250

cd/m2

DQS Predictions to Guide Display Development: TV

DQS allows developers to determine the combinations of display parameters that maximize the proportion of the population that will prefer their display. Cost, supply and other constraints can then be used in deciding how to build the most appealing display for customers

e.g. diminishing returns to resolution: look to other factors to improve quality in cost-effective manner

DQS: Mobile

D2,

P7

P7G71

6

D2,

P6

Model Display Size DQSG716 5’’ 28.3

P6 4.7’’ 28.9D2 5” 32.7P7 5‘‘ 32.7

D2

G71

6P6

, P7

P6, G

716

Isoquality Curves: Understanding Trade-Offs

Device type LCD

Diagonal 48”





Each curve represents combinations of gamut and

luminance that have the same DQS

Each curve represents combinations of gamut and

luminance that have the same DQS

Device type LCD

Diagonal 48”





Increasing distance between isoquality curves reflect saturation shown in previous slides


DQS provides the ability to understand

tradeoffs in display quality

Device type LCD

Diagonal 48”





Relatively small gamut change produces large gains when luminance is high


DQS provides the ability to understand

tradeoffs in display quality

Device type LCD

Diagonal 48”





Similarly, relatively small luminance change produces large gains at high gamut


Further Points Regarding DQS Interpretation

DQS is designed to predict display quality over a large set of images. Apparent quality is highly image dependent.

Example 1: The pair of images on the top correspond to approximately the DQS difference as the pair on the bottom but most people perceive a larger difference in the bottom two images


DQS is designed to predict display quality over a large set of images. Apparent quality is highly image dependent.

Example 2: High resolution only matters for images with fine detail. That is, increasing the resolution of a display will not improve the appearance of images with fuzzy content.

A Note About Image Processing

Image processing has become increasingly important for image compression, image enhancement and artistic control

Good image processing can influence the apparent quality of a display

However, DQS is designed only to quantify the influence of the physical parameters of the display

That is, it captures the limits of what a display can do but is silent in regard to what can be done within those limits

A Note About Image Processing: Example

Image processing can enhance image appearance within the limits set by the physical display properties. It cannot operate outside of these limits.

It can be difficult to determine whether the apparent quality of a display is a result of better image processing or superior display manufacture/design.


1. DQS is designed to predict display quality over a large set of images. Apparent quality is highly image dependent

2. It is based on side-by-side viewing

3. Does not account for form factors (e.g. may give a 10” mobile phone a high display score but this is impractical)

4. It is a work in progress that will benefit greatly from customer feedback

Future Development

DQS: Future Developments

Future development directions will be guided by feedback from users but here are some initial proposals:

1. Develop better characterization of image dependence (i.e. a set of quality scores based on image type)

2. Incorporate effects of material properties (e.g. screen gloss) and common defects (color mura, sparkle, missing pixels)

Thank You!


Disclaimer Product Use: All statements, technical information and recommendations contained in this document are based upon tests

or experience that 3M believes are reliable. However, many factors beyond 3M’s control can affect the use and performance of a 3M product in a particular application, including the conditions under which the product is used and the time and environmental conditions in which the product is expected to perform. Since these factors are uniquely within the user’s knowledge and control, it is essential that the user evaluate the 3M product to determine whether it is fit for a particular purpose and suitable for the user’s method of application.

Warranty and Limited Remedy: Unless stated otherwise in 3M’s product literature, packaging inserts or product packaging for individual products, 3M warrants that each 3M product meets the applicable specifications at the time 3M ships the product. Individual products may have additional or different warranties as stated on product literature, package inserts or product packages. 3M MAKES NO OTHER WARRANTIES, EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, ANY IMPLIED WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR ANY IMPLIED WARRANTY ARISING OUT OF A COURSE OF DEALING, CUSTOM OR USAGE OF TRADE. User is responsible for determining whether the 3M product is fit for a particular purpose and suitable for user’s application. If the 3M product is defective within the warranty period, your exclusive remedy and 3M’s and seller’s sole obligation will be, at 3M’s option, toreplace the product or refund the purchase price.

Limitation of Liability: Except where prohibited by law, 3M and seller will not be liable for any loss or damage arising fromthe 3M product, whether direct, indirect, special, incidental or consequential, regardless of the legal theory asserted, including warranty, contract, negligence or strict liability

Additional Support Slides

The Eye’s Optics

Campbell & Green, 1965; Rodeick 1998

Point Spread Function



Why does sensitivity decrease with high levels of detail?


Modulation Transfer Function

5.8 mm pupil

2 mm pupil

The Eye’s Optics

Campbell & Gubisch, 1958; Campbell & Green, 1965

Point Spread Function

The Eye’s Optics

5.8 mm pupil

2 mm pupil

Campbell & Gubisch, 1958; Campbell & Green, 1965

Contrast of fine detail is relatively reduced by eye’s optical system

Explains why sensitivity decreases with detail



Decrease in sensitivity at high levels of detail largely due to optical factors (aberrations & diffraction)


The Eye’s Optics

People cannot see images that have finer detail than 60 cycles per degree of visual angle. Why?


The Eye’s OpticsHard cut-off at ~60 cycles per degree of visual angle due to sampling (“Nyquist”) limits of the human photoreceptor mosaic

The Eye’s OpticsNyquist limit of the fovealcone sampling mosaic is ~60 cycles per degree

DQS: Current Form

Barten’s SQRI integrated influence of: Luminance, Size, Resolution, Contrast

Influence of color gamut on display quality

s

Thank you

3M’s Display Quality Score · 3M’s Display Quality Score 3M’s tool to guide the development...

Documents

Transcript of 3M’s Display Quality Score · 3M’s Display Quality Score 3M’s tool to guide the development...