Japan Advanced Institute of Science and Technology

Japan Advanced Institute of Science and Technology

JAIST Repository

https://dspace.jaist.ac.jp/

Title

デル

Author(s)

Citation

Issue Date

Type

Text version

URL

Rights

Description

Coded Images Considering Picture Quality Class

FURUSHO Yoshiko

Scho ol of Information Science,

Japan Advanced Institute of Science and Technology

February 13, 1998

Keywords: picture quality evaluation, color dierence,JPEG coding, objective

quality evaluation, picture quality class.

1 Introduction

High compressible co ding, for example JPEG, bring about block distortion, disconti-

nuity of outline, blurry color and so on. In other to evaluation of coded images, the

subjectiveassessmenthas b een used. Howeverthe assessment needssubject'shard labor

and a lot of time. If the evaluation mo del considering human sight and perception trait

is constructed,the aboveproblems will be solved.

This papershowsthe picturequalitymodelfor evaluationof JPEGcodedcolorimage

onRGB,YCbCr,CIEL*a*b*andCIEL*u*v*colorspace. Thedistortionmo delsconsist

of fundamental quality factors dened by color dierence on variety color space. The

equations of evaluation are obtained by Multivariate Analysisof distortion models. The

goal of this research make clear on suit color space for picture quality evaluation, to

construct an evaluation model considering picture quality class and human observing

view p oint,and to improve the precision of model.

2 Picture Quality Evaluation Model on Variety Color

Space

The dierence b etween the original and the reproduced signal is dened as the color

dierence which is on the RGB, YCbCr, CIE L*a*b*, CIE L*u*v* color space. Funda-

mentaldistortionfactors aredenedbythefunctionofcolordierence,andthe equations

Copyrightc 1998byFURUSHOYoshiko

are fundamental distortion factors.

F

1

: Themean of color dierence.

F

2

: Errorsat the co ding sub-block boundaries.

F

3

: Acoecientof auto correlation of errors.

F

4

: Errorsat the neighb orho odof outline.

Evaluation Model Considering Picture Quality Class

The evaluation model of picture quality evaluation model for color coded images con-

sidering picture quality class is constructed improve precision. Co ding image quality is

classifyas a highqualityclass and a lowqualityclass. Indispensable distortion factors is

make clear ineach evaluation model.

Measurement of observing point at subjective assessment

In case of that error has a spatial relation namely block distortion and discontinuity

of outline, humanp erceive then ten times asmuchas the random noise. I construct the

evaluation model based onobserving viewpoint. For that reason I measureof observing

point at subjective assessment of the quality of color coded images. Therefore I make

clearobservingviewp ointinassessmentand contributetomakeadecisionforassessment

score.

Apply to observing point for the evaluation model of picture

quality

Fundamentaldistortion factorsF

4

has aproblem. Itdoesnotmeasuretheforapp earance

discontinuity outline in low quality class. New evaluation model considering observing

view p ointdenes new F

4

to measurefor observing viewpoint's error.

3 Conclusion

I constructed of evaluation model of picture quality on RGB, YCbCr, CIE L*a*b*, and

CIE L*u*v* color space. The luminance-chrominance separated color system suit to use

evaluation. The best color space is uniform color space. One of the most suit able color

space is CIE L*u*v*. This mo del evaluates the picture quality with high accuracy and

estimates Mean Opinion Score well(87% accuracy).

However, this mo del is not precise not in low quality class. Therefore I construct of

evaluation model considering picture quality class on CIE L*a*b* color space. As the

result of construct of evaluation model improvedthe precision.

is vary from picture quality class. F

3

is an indispensable distortion of lowquality class.

It is anautocorrelation co ecientof errors.

Large distortion of image, for example texture pattern, do not take shape, however

neighb orhoodofoutlineerrorisnotableerrorinhigh qualityclass. F

4

isanindispensable

distortion of high quality class. Itis a measureforerrors at the neighb orhood of outline.

Measurementofobservingpointatsubjectiveassessmentofthe qualityof colorcoded

images one account of indispensable distortion factors is varyfrom picture quality class.

Observingp ointisvaryfromexperienceofassessmentp ossessionandp eculiaritinpicture

quality. Thereareobservingpointsonlargeimagedistortion,forexampleblockdistortion,

in low quality class and there are it onoutline in high quality class. Apply to observing

pointforevaluationmodel ofpicturequality,butthis mo del evaluatesthepicturequality

with mid accuracy and estimates Mean Opinion Score well (70% accuracy). To practice

the assessmentfor many parsonwill improvethe picture quality evaluation .