12 Piotr Didyk 2 Tobias Ritschel 3 Elmar Eisemann 3 Karol Myszkowski 2 HansPeter Seidel 2 Apparent Resolution Enhancement for Animations 1 University of Wroc ław Poland ID: 285633
Download Presentation The PPT/PDF document "Krzysztof Templin" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Krzysztof Templin1,2 Piotr Didyk2 Tobias Ritschel3Elmar Eisemann3 Karol Myszkowski2 Hans-Peter Seidel2
Apparent Resolution Enhancement for Animations
1
University of
Wroc
ław, Poland
2
MPI Informatik, Germany
3
T
élécom
ParisTech
, FranceSlide2
Motivation
easily
~
50
MPix
~
2-8
MPix
1px
→
~
9
receptorsSlide3
Standard methods
Cropping
DownsamplingSlide4
Decomposition into subframes
high-resolution
image
low-resolution
subframes
decompose
perceived high-resolution
image
integrateSlide5
Related work“Display
Supersampling”[
Damera-Venkata
and Chang 2009]
multiple projectors, one subframe each
“
Wobulation
: Doubling the Addressed Resolution of Projection Displays”
[Allen and
Ulichney
2005]
single projector,
two
subframes, subpixel shift
“Apparent Display Resolution Enhancement for Moving Images”
[Didyk et al. 2010]
multiple subframes moving over
120Hz LCD displaySlide6
Didyk et al.
timeSlide7
Didyk et. alSlide8
Didyk et. alSlide9
p
ixel
1
p
ixel
2
frame 1
frame 3
frame 2
p
ixel
1
p
ixel
2
Temporal
d
omainSlide10
B
C
A
A
B
C
p
ixel
1
p
ixel
2
receptor
frame 1
frame 3
frame 2
temporal integration
Temporal
d
omain – static caseSlide11
A
B
C
p
ixel
1
p
ixel
2
frame 1
frame 3
frame 2
temporal integration
Temporal
d
omain – dynamic case
receptor
B
A
CSlide12
Slide13
receptor
Receptor signal:
–
segment
–
pixel in segment
i
–
intensity of pixel
x
in segment
i
–
weights
proportional to the length of the
segment
Temporal integration modelSlide14
Receptors at grid points. Perfect tracking. Receptor layoutSlide15
prediction for
one receptor
Prediction in
e
quations
subframes
retina image
integration
modelSlide16Slide17
integration
model
Optimization
p
roblem
subframes
high-resolution imageSlide18Slide19
Panning (integer motion)
1
2
3
1’
2’
3’Slide20
Critical Flicker Frequency
Critical Flicker Frequency – Hecht and Smith’s data fromBrown J.L. Flicker and Intermittent Simulation
10 Hz
20 Hz
30 Hz
40 Hz
5
0 Hz
6
0 Hz
-3
-1
1
3
-3
Temporal contrast
Frequency
Three-frame cycleon 120 Hz display
40 Hz signal
Fusion frequency depends on:
Temporal contrastSpatial extent
19
deg
1 deg
0.3
degSlide21
Non-integer motion
1
2
3
4
5
6Slide22
1
2
3
4
5
6
Non-integer motionSlide23
Non-integer motion
?
≈
?
1
2
3
4
5
6Slide24
General animationsMotion already present – no need to move.
Eye follows the motion of the corresponding detail.Local optimization, in space and time.Slide25
Receptors paths
Problem
:
too sparse
non-uniform distribution
Solution: we reintroduce receptorsSlide26
Receptors paths
Solution: we reintroduce receptorsSlide27
Optimization
retina image
current solution
original
subtract
error
integrate
project back
improved solution
iterate
rows / s = 120 × resolution × lifetime
optimal
subframesSlide28
GPU implementation
simple fragment
shader
line drawing with
alpha blendingSlide29
Lanczos filteringStandard approach: radius 6.Smaller kernels leave aliasing in frames.Can integrate, similarly to optimized solution.We compare to radius 3, 4, 5 and 6.Similar to [Basu and Baudisch 2009].No perfect solution [Mitchell and Netravali 1988].Slide30
Results (general animations)More detailed than Lanczos 6.Details similar to Lanczos 4, but less aliasing.Slide31
Perceptual studyNumber of participants: 14.Two-step procedure: 1. Lanczos kernel adjustment. 2. Lanczos vs. ours comparison.Question asked: which reproduces the original better.
Study showed, that our method gives the best results:
Method
Preference
Lanczos
3
1%
Lanczos 4
3%
Lanczos 5
17%Lanczos 619%
Our
60%Slide32
Velocity vs. QualitySubframes integrate giving impression of increased resolution.Often fusion is not complete – some artifacts visible.But not always.Slide33
ConclusionWe generalized previous results, and showed how to enhance depiction of details in arbitrary animations.Compared our algorithm to other filtering methods in a perceptual study.Designed an efficent GPU implementation.Slide34
Future workHigher refresh rates.Flicker reduction methods.Faster implementation.Eye-tracking.Non-uniform sampling.Other media.Slide35
Thank you!Apparent Resolution Enhancement for
AnimationsKrzysztof Templin Piotr Didyk Tobias Ritschel
Elmar Eisemann Karol Myszkowski Hans-Peter Seidel
http://www.mpi-inf.mpg.de/resources/ResolutionEnhancement/Animations/