P. 2May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Image quality measurement:
real world challenges
Norman Koren Imatest LLC Boulder, Colorado
www.imatest com
• Background: Predicting image
quality; Imatest structure
• Image quality factors and how
they are measured
• Imatest modules: review
P. 3
Image quality example
May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
DOC/NIST building in Boulder (atomic clock): Street view boundary
Poor sharpness, smudged shadow detail, JPEG artifacts, flare light(?)
Imatest was created to
predict imaging system performance.
Lost tonal detail
Sharpness
Flare light?
JPEG artifacts
P. 4May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest: Basics
• Photograph test chart (standard or user-created) in controlled
environment or as part of a scene.
• Cannot separate lens, sensor, signal processing.
• Analyze image for relevant quality factor.
P. 5May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest: Background
• Created to enable individual photographers to
test lenses and cameras.
– Lens sharpness?
– Camera dynamic range?
– Color reproduction accuracy?
• Widely adopted by industry: mobile imaging
and many others.
• Compiled Matlab.
• Downloaded from www.imatest.com.
• Modules analyze images of standard targets.
P. 6May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Tour of image quality factors
with examples of Imatest analysis
Monument Valley/Hunt’s Mesa image illustrates
image quality factor degradations.
Original
on
left
Degraded
on
right
Issues to think about:
• Capture vs. post-processing
• Objective measurements vs. subjective
judgment
P. 7May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Module summary Image quality factor
Sharpness
Color accuracy
Tonal response
and contrast
Dynamic range
Exposure accuracy
Uniformity
Lens distortion
Lateral chromatic
aberration
Noise
Colorcheck,
Multicharts
SFR (Spatial fre-
quency response)
Stepchart
Distortion
Light Falloff
Log frequency-
Contrast
Loss of detail (NR)
Module
Lens flare
P. 8May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Spatial frequency response (SFR);
Modulation transfer function (MTF)
• Upper: Sine and bar patterns:
original and blurred.
• Middle: Level of the blurred bar
pattern (red curve). Contrast
decreases at high spatial
frequencies.
• Lower: the corresponding MTF
(SFR) curve (blue curve).
• Low frequency MTF is defined
to be 1 (100%). MTF can be
larger than 1.
• Strongly affected by signal
processing (sharpening).
Measuring sharpness
Spatial frequency
P. 9May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Sharpness (loss)
• Arguably the most important factor
• Determines how much detail can be conveyed
• Affected by the lens, sensor, and digital signal
processing (sharpening)
• Measured by Spatial frequency response (SFR),
AKA Modulation Transfer Function (MTF)
SFR (Spatial fre-
quency response)
Log frequency-
Contrast
original | blurred
P. 10May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Sharpening
• Most digital images look soft without
sharpening.
• Subtracts a fraction of neighboring pixels from
each pixel.
• Boosts contrast & MTF at high spatial
frequencies.
• Applied to virtually all digital camera images—
in the camera, RAW converter, and/or image
editor.
• Different amounts of sharpening in different
cameras makes comparisons challenging.
Transfer function: MTFsharp( f ) = (1 – ksharp cos(2π f V ))/(1-ksharp)
where V = Sharpening radius / pixel spacing
Black– unsharpened
Red: sharpened
P. 11May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Sharpness (oversharpening)
• Too much digital sharpening causes severe
“halos” at edges.
SFR (Spatial fre-
quency response)
• Peak in MTF
response.
• Boosts MTF50.
• Common in
compact digital
cameras.
• Looks OK in
small images;
bad enlarged.
Log frequency-
Contrast
original | oversharpened
P. 12May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
SQF: Subjective Quality Factor
A measure of subjective (perceptual) sharpness,
optionally displayed in SFR, that combines
• MTF,
• The human eye’s
Contrast sensitivity
function (CSF) (peaks at
6-8 cycles/degree),
• Image height,
• Viewing distance
A+ A B+ B C+ C D F
94-100 89-94 84-89 79-84 69-79 59-69 49-59 Under 49
CSF of the human eye
P. 13May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Lateral chromatic aberration
• Seen as “color fringing” near corners.
• Can be digitally corrected.
SFR (Spatial fre-
quency response)
original | with CA
P. 14May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Noise
• A serious degradation; corresponding to
grain in film.
• Software noise reduction can remove fine
detail.
Colorcheck
Stepchart
original | noisy
P. 15May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Color accuracy
• Uses the GretagMacbeth ColorChecker (in
Colorcheck; other charts in Multicharts).
• Errors displayed in L*a*b* space.
Colorcheck
original | color-shifted
• Several color
difference
metrics can be
selected.
• Reference
colors can be
selected to be
accurate or
pleasing.
P. 16May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Tonal response and Contrast
• Pixel level ≈ luminanceγ (γ is gamma =
contrast); S-curve often superimposed.
• Image contrast is gamma (γ) in mid-tones.
Stepchart
original | clipped
P. 17May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Dynamic range
• The exposure range a
camera can record at a
quality level specified by max
noise or min SNR. Units of f-
stops.
• Reflected step charts have
insufficient range; trans-
mission chart recommended
(Stouffer T4110 with Dmax =
4.0 shown).
Stepchart
P. 18May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Lens flare (veiling glare)
• Stray light bouncing between lens elements
or off lens barrel (interior). Important when
lighting is uncontrolled.
• Overall fogging of image (loss of shadow
detail): can be measured
• “Ghost” imaging: difficult to measure,
predict.
• Measured using a
“black hole” (cavity)
with white surround
next to step chart.
• Veiling glare = V = 100% (L(black hole) /
L(white surface))
Stepchart
original | with flare
P. 19May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Exposure accuracy
• Important in cameras with auto-
exposure
• Affected by history: may change
after exposure to very bright or
dim light.
• Calculated from reference
values for step chart or
ColorChecker and gamma (γ).
Stepchart
Colorcheck
original | overexposed
P. 20May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Light falloff
• Measures light falloff due to lens and
sensor, as well as color shifts due to “pixel
shading” at sensor.
Light Falloff
original | vignetted
P. 21May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Lens distortion
• Can be measured using a grid or a single
line near the image boundary.
• Several correction coefficients calculated.
Distortion
original | barrel distortion
P. 22May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Loss of detail from
Signal processing (ISO 80)
Measure contrast loss as a function of
spatial frequency and chart contrast.
boundary.
Log frequency-
Contrast
Compact digital camera, ISO 80
original | lost detail
P. 23May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Loss of detail from
Signal processing (ISO 800)
More noise reduction at higher ISO speed
results in more contrast loss at high spatial
frequencies, especially at lower contrasts.
Log frequency-
Contrast
Compact digital camera, ISO 800
Noise artifacts
Test chart
P. 24May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest modules: Stepchart
Photograph a step chart:
reflective (Kodak Q-13/Q-14, etc.),
transmission (Stouffer T4110, etc.)
Measure
• Tonal response,
• Gamma (contrast; average and
instantaneous),
• Noise (or SNR),
• Dynamic range (transmission
charts only),
• Exposure error (reflective charts
only).
P. 25May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest modules: Colorcheck
Photograph the GretagMacbeth
Color checker.
Measure
• Color accuracy
(various lighting
conditions),
• Tonal response,
• Gamma,
• Noise,
• Exposure error.
P. 26May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest modules: Multicharts
Photograph the test chart (many charts are
supported). Highly interactive interface.Measure
• Color accuracy
(various lighting
conditions),
• Tonal response,
• Gamma.
P. 27May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest modules:
Distortion
Photograph a grid.
• 3rd and 5th order,
• Tangent/arctangent,
• SMIA TV distortion.
Measure Distortion and correction
coefficients in several models:
Light Falloff
Photograph a uniform white
or gray region.
• Light falloff (vignetting; uniformity),
• Sensor noise detail,
• Dead and hot pixels.
Measure
P. 28May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest modules: SFR
(Spatial frequency response: SHARPNESS)
Photograph a slanted-
edge target. Can be
printed on a high quality
inkjet printer or be a part
of the ISO 12233 chart.
Measure
• Average edge response (upper
plot); 10-90% rise distance,
• SFR (spatial frequency response
= MTF); MTF50 (an excellent
metric for image sharpness)
• Lateral chromatic aberration.
Dashed red lines (- - -) are for
standardized sharpening.
P. 29May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Imatest summary
Sharpness
Noise
Dynamic range
Color accuracy
Image quality is determined by several factors.
Imatest analyzes
Light falloff
Lens distortion
Flare light
Data compression loss
Tonal response and
contrast
Lateral chromatic
aberration
Exposure accuracy
• Some more affected by capture; others by post-processing.
• Many can be improved with post-processing.
• Weighting of each factor depends on individual preference,
application.
• Difficult to define a single measure of image quality.
P. 30May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Oversharpened image
5 MPXL compact digital camera
Peaks in both domains. “Halo”
(overshoot) at the edge.
MTF50 is unrealistically high.
P. 31May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Undersharpened image
11 MPXL DSLR
Edges are rounded; no overshoot.
Image can benefit from additional
sharpening.
MTF50 is lower than it would be
with a reasonable amount of
sharpening. MTF50 (LW/PH) is
lower than the 5 MPXL camera.
It is difficult to make a
fair comparison
between under- and
oversharpened images.
P. 32May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Standardized sharpening I
Algorithm:
Sharpen (or de-sharpen) the image with a
fixed radius R (usually between 1 and 2;
the value used in most compact digital
cameras) so MTF at f = 0.3 * Nyquist (0.15
cycles/pixel) is equal to MTF at f = 0.
Standardized sharpening is a
strategy for comparing camera
performance in the presence
of differences in sharpening.
The response with standardized
sharpening is shown by the dashed
(– – –) red curves.
MTF50(corr) indicates sharpness.
P. 33May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Standardized sharpening II
Standardized sharpened edges have a
small amount of overshoot: typical for
manual sharpening.
Thus undersharped 11 MPXL DSLR
image has been Standardized
sharpened with R = 2. MTF50 (LW/PH)
is higher than the 5 MPXL camera.
( )
sharp
origscansharp
stdsh
k
fMTFdRfk
fMTF
!
!
=
1
)()/2cos(1
)(
"
)(/1)2cos(
)(/11
eqlorigeql
eqlorig
sharp
fMTFfR
fMTF
k
!
!
=
"
where
Standardized sharpening equations:
P. 34May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Standardized sharpening radius R
Same pulse (11 MPXL DSLR),
Standardized sharpened with R = 1:
Sharper (higher MTF50) than R = 2.
R = 1 often gives best results for
undersharpened DSLRs.
Larger R is appropriate systems with
poor sharpness (low MTF50).
R = 2 usually works better for de-
sharpening compact digital cameras,
typically (over)sharpened with R = 2.
No general algorithm
for selecting R
P. 35May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Standardized sharpening
Conclusions
• Useful for comparing performance of different
cameras with light to moderate signal processing.
but
• Can be fooled by sophisticated signal processing
(complex sharpening & noise reduction with
thresholds)
– Almost any response curve can be replicated with sufficient
sharpening, but...
– Excessive sharpening boosts noise & other artifacts. Can
worsen appearance.
• Additional tests needed.
P. 36May. 18, 2006 Imatest : introduction
Norman Koren www.imatest.com
Slanted-edge measurements
Conclusions
• Excellent for measuring sharpness:
Convenient, accurate, valid well beyond Nyquist frequency.
• Affected by in-camera sharpening. Best with no or known
sharpening.
• Cannot distinguish between response above Nyquist
caused by weak anti-aliasing filter and sharpening; does
not indicate potential seriousness of Moiré fringing.
• Additional tests will be added to fully characterize system
quality, e.g., CIPA DC-003 & Siemens Star, which can
measure the onset of aliasing.
本文档为【PhotoTech_26_Imatest_Slides】,请使用软件OFFICE或WPS软件打开。作品中的文字与图均可以修改和编辑,
图片更改请在作品中右键图片并更换,文字修改请直接点击文字进行修改,也可以新增和删除文档中的内容。
该文档来自用户分享,如有侵权行为请发邮件ishare@vip.sina.com联系网站客服,我们会及时删除。
[版权声明] 本站所有资料为用户分享产生,若发现您的权利被侵害,请联系客服邮件isharekefu@iask.cn,我们尽快处理。
本作品所展示的图片、画像、字体、音乐的版权可能需版权方额外授权,请谨慎使用。
网站提供的党政主题相关内容(国旗、国徽、党徽..)目的在于配合国家政策宣传,仅限个人学习分享使用,禁止用于任何广告和商用目的。