Rochester Institute of Technology RIT Scholar Works Presentations and other scholarship 10-18-1999 Measuring and Modeling Image Quality Mark Fairchild Rochester Institute of Technology Follow this and additional works at: hp://scholarworks.rit.edu/other is Presentation is brought to you for free and open access by RIT Scholar Works. It has been accepted for inclusion in Presentations and other scholarship by an authorized administrator of RIT Scholar Works. For more information, please contact [email protected]. Recommended Citation Fairchild, M.D. (1999). Measuring and modeling image quality [PowerPoint slides]. Presented at the Center for Imaging Science Industrial Associates Meeting, Rochester, NY., 18 October.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Rochester Institute of TechnologyRIT Scholar Works
Presentations and other scholarship
10-18-1999
Measuring and Modeling Image QualityMark FairchildRochester Institute of Technology
Follow this and additional works at: http://scholarworks.rit.edu/other
This Presentation is brought to you for free and open access by RIT Scholar Works. It has been accepted for inclusion in Presentations and otherscholarship by an authorized administrator of RIT Scholar Works. For more information, please contact [email protected].
Recommended CitationFairchild, M.D. (1999). Measuring and modeling image quality [PowerPoint slides]. Presented at the Center for Imaging ScienceIndustrial Associates Meeting, Rochester, NY., 18 October.
RITRIT Munsell Munsell Color Science Laboratory Color Science Laboratory
Measuring and ModelingMeasuring and ModelingImage QualityImage Quality
Center for Imaging Science Industrial Associates MeetingCenter for Imaging Science Industrial Associates Meeting
October 18, 1999October 18, 1999
OverviewOverview
•A Proposed Approach to Measuring Overall•A Proposed Approach to Measuring OverallImage QualityImage Quality
(Ideas & Plans; Not Results)(Ideas & Plans; Not Results)
•An Invitation for Comments, Suggestions,•An Invitation for Comments, Suggestions,Data, Assistance, Funding, Data, Assistance, Funding, Etc.Etc.
OutlineOutline
•What is Image Quality?•What is Image Quality?
•Previous Approaches•Previous Approaches
•Overview of a Proposed Approach•Overview of a Proposed Approach
•Ongoing Research•Ongoing Research
•Future Questions•Future Questions
(Focus is on Static Images; Temporal Aspects Ignored for Today)(Focus is on Static Images; Temporal Aspects Ignored for Today)
What is Image Quality?What is Image Quality?
Reductions in image quality correspond toReductions in image quality correspond toperceptible visual perceptible visual differencesdifferences from some from someideal and the ideal and the magnitudemagnitude of such differences. of such differences.
Image Quality as an Interval ScaleImage Quality as an Interval Scale
Interval scales have no meaningful zero.Interval scales have no meaningful zero.(What is zero image quality?)(What is zero image quality?)
Reduction in image quality can be a ratio scale.Reduction in image quality can be a ratio scale.(Zero reduction in IQ is a sub-threshold difference.)(Zero reduction in IQ is a sub-threshold difference.)
Meaningful Range of ValuesMeaningful Range of Values
ReductionReductionin I.Q.in I.Q.
DifferenceDifferenceMetricMetric
Measurable
Meaningful
Detectable
Some ApproachesSome Approaches
Generally, various dimensions of image qualityGenerally, various dimensions of image qualityare treated individually while either ignoring,are treated individually while either ignoring,or holding constant, other dimensions.or holding constant, other dimensions.
Limitations of Previous ApproachesLimitations of Previous Approaches
•Color Approaches Tend to Ignore Spatial Attributes•Color Approaches Tend to Ignore Spatial Attributes
EE*=*=50 for each pixel50 for each pixelV = 0 for appropriate viewing distanceV = 0 for appropriate viewing distance
•Spatial Approaches Tend to Ignore Color Appearance•Spatial Approaches Tend to Ignore Color Appearance(Luminance Only, No Adaptation, etc.)(Luminance Only, No Adaptation, etc.)
•Image Specifications Tend to Ignore Human Perception•Image Specifications Tend to Ignore Human Perception
A Unified ApproachA Unified ApproachAn image quality metric can be derived as a measureAn image quality metric can be derived as a measure
of perceived difference from an ideal image.of perceived difference from an ideal image.
S.N.S.N. Pattanaik Pattanaik, J.A., J.A. Ferwerda Ferwerda, M.D. Fairchild, and D.P. Greenberg,, M.D. Fairchild, and D.P. Greenberg,AA multiscale multiscale model of adaptation and spatial vision for image model of adaptation and spatial vision for imagedisplay, display, Proceedings of SIGGRAPH 98,Proceedings of SIGGRAPH 98, 287-298 (1998). 287-298 (1998).
S.N.S.N. Pattanaik Pattanaik, M.D. Fairchild, J.A., M.D. Fairchild, J.A. Ferwerda Ferwerda, and D.P. Greenberg,, and D.P. Greenberg,MultiscaleMultiscale model of adaptation, spatial vision, and color model of adaptation, spatial vision, and colorappearance, appearance, IS&T/SID 6th Color Imaging Conference,IS&T/SID 6th Color Imaging Conference, Scottsdale, Scottsdale,2-7 (1998).2-7 (1998).
•A Spatial Vision Model that Incorporates•A Spatial Vision Model that IncorporatesColor-Appearance ConceptsColor-Appearance Concepts
G.M. Johnson, and M.D. Fairchild,G.M. Johnson, and M.D. Fairchild, Full-spectral color Full-spectral colorcalculations in realistic image synthesis, calculations in realistic image synthesis, IEEE ComputerIEEE ComputerGraphics & ApplicationsGraphics & Applications 19:419:4, 47-53 (1999)., 47-53 (1999).
G.M. Johnson and M.D. Fairchild,G.M. Johnson and M.D. Fairchild, Computer synthesis of Computer synthesis ofspectroradiometricspectroradiometric images for color imaging systems images for color imaging systemsanalysis, analysis, IS&T/SID 6th Color Imaging Conference,IS&T/SID 6th Color Imaging Conference, Scottsdale, Scottsdale,150-153 (1998).150-153 (1998).
•Synthesis of Spectral Images•Synthesis of Spectral Images•More Realistic Test Targets•More Realistic Test Targets
A.A. Vaysman Vaysman and M.D. Fairchild, and M.D. Fairchild, Degree of Degree of quantization quantization and andspatialspatial addressability addressability trade- trade-offsoffs in perceived quality of in perceived quality ofcolor images, color images, Color Imaging: Device Independent Color, ColorColor Imaging: Device Independent Color, ColorHardcopy, and Graphic Arts III,Hardcopy, and Graphic Arts III, Proc Proc. SPIE. SPIE 3300, 250-261 3300, 250-261(1998).(1998).
•Image Quality as a Function of Bits/Pixel and DPI•Image Quality as a Function of Bits/Pixel and DPI
•Construction of Very-High Quality•Construction of Very-High QualityTest Targets Test Targets (~1.3GB/Image)(~1.3GB/Image)
•Creation of New Techniques•Creation of New Techniques((e.g.,e.g., Ray Tracing in addition to OpenGL) Ray Tracing in addition to OpenGL)
•Ongoing Refinement of Procedures•Ongoing Refinement of Procedures
•Bigger, Faster Computer•Bigger, Faster Computer
Garrett JohnsonGarrett Johnson
3. Visual Quality Scale3. Visual Quality Scale
•Start with Synthetic Image•Start with Synthetic Image
• Degrade Along Several Typical Dimensions• Degrade Along Several Typical Dimensions(dpi, (dpi, bppbpp, color , color responsiivityresponsiivity, tone , tone reporductionreporduction, , etcetc.).)
•Visually Scale Quality•Visually Scale QualityPaired Comparison w/AnchorPaired Comparison w/AnchorSingle Viewing Condition for NowSingle Viewing Condition for Now
•Bigger, Faster Computer•Bigger, Faster Computer
Garrett JohnsonGarrett Johnson
Putting It All TogetherPutting It All Together
•Measures from •Measures from MihaiMihai
•Scales from Garrett•Scales from Garrett
•Does it Work??•Does it Work??
Future QuestionsFuture Questions
•Improved Vision Model / Other Vision Models?•Improved Vision Model / Other Vision Models?(A Simpler Approach?)(A Simpler Approach?)
•More Extensive Image Degradations•More Extensive Image Degradations(Real and Simulated)(Real and Simulated)
•More Visual Data•More Visual Data
•Iterate•Iterate
1. Vision Model1. Vision Model
Could other models work as well or better?Could other models work as well or better?
Is the complexity justified?Is the complexity justified?(S-CIELAB approach enough?)(S-CIELAB approach enough?)
What is the optimum combination?What is the optimum combination?(Analogous to color appearance model evolution?)(Analogous to color appearance model evolution?)
•Collection of Data from Other Sources•Collection of Data from Other Sources
•What Works, What Sells, •What Works, What Sells, EtcEtc..
How Can You Help?How Can You Help?
•Contribute Data•Contribute Data
•Point Out Apparent Flaws•Point Out Apparent Flaws
•Suggest Other Models to Try•Suggest Other Models to Try
•Comment on Utility of the Answer•Comment on Utility of the Answer
•Suggest Enhancements to This Approach•Suggest Enhancements to This Approach
•Fund a Research Project!•Fund a Research Project!
•Students: Join in on this Research!•Students: Join in on this Research!
AcknowledgementsAcknowledgements
Fuji Photo-Film Fuji Photo-Film (Kazuhiko (Kazuhiko TakemuraTakemura))Supporting research on vision model and image qualitySupporting research on vision model and image quality
scaling.scaling.
Eastman Kodak Eastman Kodak (Paula (Paula AlessiAlessi))Supporting research on spectral image synthesis andSupporting research on spectral image synthesis and
spectral portraiture.spectral portraiture.
How Can You Help?How Can You Help?
•Contribute Data•Contribute Data
•Point Out Apparent Flaws•Point Out Apparent Flaws
•Suggest Other Models to Try•Suggest Other Models to Try
•Comment on Utility of the Answer•Comment on Utility of the Answer
•Suggest Enhancements to This Approach•Suggest Enhancements to This Approach
•Fund a Research Project!•Fund a Research Project!
•Students: Join in on this Research!•Students: Join in on this Research!