Introduction to Image Quality Metrics Larry N. Thibos School of - - PowerPoint PPT Presentation

OSA Webinar for Metrology 01 July 2015

Introduction to Image Quality Metrics

Larry N. Thibos School of Optometry, Indiana University, Bloomington, IN 47405 thibos@indiana.edu

Begin with a definition Key issues: How will the image be used, and for what purpose? Toraldo di Francia G. “Modern trends in the evaluation of

• ptical images.” J Opt Soc Am. 1957;47(6):507.

Which image has better quality? Focus near Focus far

Some common uses of images Natural images record the appearance of real objects

• Biological perception and action

– Direct imaging of physical objects by the eye onto the retina – Indirect imaging by a man-made imaging system (e.g. a camera), which becomes a secondary object imaged by the eye

• Machine vision for controlling machines (e.g robots)

Artificial images render imaginary objects

• Computer-to-human communication
• Computer-to-computer communication

A paradigm for determining perceived visual quality

Which image would you share on social media?

A method for developing a metric of image quality Object Imaging system #1 Imaging system #2 Image #1 Image #2 Observer decides Internal criterion What measure of the imaging system predicts human preferences (independent of object)?

Empirical measurement of perceived image quality Using the observer’s intuitive sense of image quality,

• Observers prefer images with the largest number of

perceived gray levels (JNDs, just-noticeable differences)

– Granger & Cupery (1972, at Kodak, Inc. with photographic prints) – Barten (1987, at Phillips, Inc. with video displays)

• Number of gray levels can be predicted from optical

characteristics of the system that produced the image. => Perceptual quality can be quantified by simple measures of the MTF

– SQF, subjective quality factor (Granger & Cupery, 1972) – SQRI, square root integral (Barten, 1987)

Objective prediction of perceived image quality Area = SQRI Barten (1987)

3 12

Area = SQF Granger & Cupery (1972)

Nonlinear axes provide a perceptually uniform space for graphical analysis.

Limitations of perceived image quality

Perceived image quality is a subjective judgment

• f appearance and esthetic appeal, not

necessarily related to the performance of a task. Basing image quality on performance of a task takes into account the purpose of the image.

Better subjective quality may not give better performance

MTF MTF

Performance image quality Equal visual performance <=> equal image quality If two images yield equal performance on a visual task, then (by definition) they have equal quality. Conversely, if two images have equal quality, then the

• ptical systems that produced the images will enable

equal performance of a task by human or machine visual systems. This generic approach to measuring image quality quantifies the imaging system that produced the images, rather than the images per se.

A performance-based method for measuring images Object Imaging system #1 Imaging system #2 Image #1 Image #2 Observer task Equal performance => equal image quality What measure of these filters reveals their equality?

Specifying optical filters Wavefront aberration map (pupil plane) Optical Transfer Function Point-spread Function

In what way are these two imaging systems the same? Filter #1 Filter #2

C2

0 = +0.6, C4 0 = +0.1

C2

0 = -0.3, C4 0 = +0.1

Spatial metrics of PSF quality*

Narrow, compact PSF => quality

• ptics.

*R. Bracewell “The Fourier Transform and Its Applications” McGraw-Hill; 1978.

Image Quality Metrics for Point Objects

Spatial Compactness

• 1. Area catching 50% light
• 2. Equivalent width
• 3. Second moment
• 4. Half-width at half-height
• 5. Correlation width

Image Contrast

• 1. Strehl ratio
• 2. Light in diffraction core
• 3. StdDev of light intensity
• 4. Entropy

Point image

Compact, high-contrast point image => quality optics.

Hi-Q Low-Q

Image Quality Metrics for Grating Objects

• 1. Cutoff frequency, rMTF
• 2. Area between rMTF, visual thresh
• 3. Cutoff frequency, rOTF
• 4. Area between rOTF, visual thresh
• 5. Strehl ratio (OTF)
• 6. Strehl ratio (MTF)
• 7. OTF volume/ MTF volume

Grating image

High contrast image w/o phase shifts => quality optics.

Hi-Q Low-Q

Taking the image consumer into account

Including characteristics of the image consumer (human or machine) in metric calculations emphasizes image components that are most useful for task performance. For example, modifying the definition of Strehl Ratio by weighting the OTF by the

• bserver’s contrast sensitivity function de-

emphasizes the less visible components of an image.

= Optical PSF Visual weight Visual PSF Optical OTF Visual sensitivity Visual OTF = X

Visual Strehl Ratio: a measure of visual quality

Compare peak of visual PSF to ideal case Compare volume of visual OTF to ideal case

Bibliography

Bracewell RN. The Fourier Transform and Its Applications. second ed. New York: McGraw-Hill; 1978. Wetherell WB. The calculation of image quality. In: Shannon RR, Wyant JC,

• editors. Applied Optics and Optical Engineering. New York: Academic Press;
• 1980. p. 172-315.

Shannon RR. The Art and Science of Optical Design. Cambridge: Cambridge University Press; 1997. (chapter 4) Martens JB. Multidimensional modeling of image quality. Proc IEEE. 2002;90(1):133-53. Thibos LN, Hong X, Bradley A, Applegate RA. Accuracy and precision of

• bjective refraction from wavefront aberrations. Journal of Vision.

2004;4(4):329-51. (Appendix defines 31 metrics)

The end

Vision Research at

http://www.opt.indiana.edu