Noise-Optimal Capture for High Dynamic Range Photography
Samuel W. Hasinoff,
Frédo Durand,
and
William T. Freeman
Publications
Samuel W. Hasinoff, Frédo Durand, and William T. Freeman,
Noise-Optimal Capture for High Dynamic Range Photography.
Proc. 23rd IEEE Conference on Computer Vision and Pattern Recognition,
CVPR 2010, pp. 553-560
[pdf]
Abstract
Taking multiple exposures is a well-established approach
both for capturing high dynamic range (HDR) scenes and
for noise reduction. But what is the optimal set of photos
to capture? The typical approach to HDR capture uses a
set of photos with geometrically-spaced exposure times, at
a fixed ISO setting (typically ISO 100 or 200). By contrast,
we show that the capture sequence with optimal worst-case
performance, in general, uses much higher and variable
ISO settings, and spends disproportionately longer capturing
the dark parts of the scene. Based on a detailed model
of noise, we show that optimal capture can be formulated
as a mixed integer programming problem.
Compared to typical HDR capture,
our method lets us achieve higher worst-case SNR in the
same capture time (for some cameras, up to 19 dB improvement
in the darkest regions), or much faster capture for the
same minimum acceptable SNR. Our experiments demonstrate
this advantage for both real and synthetic scenes.
Supplementary material
- Extension with dark current noise, discussion of high vs.
varying ISO
[pdf]
- Poster presented at CVPR 2010
[pdf]
- Invited tutorial at ICCP 2010, Fundamentals of Computational Photography: Sensors and Noise
[pptx]
Synthetic example: Nancy church (Figs. 4-6)
- source HDR photo courtesy Rafał Mantiuk
[CC3.0 license]
- dynamic range of 33900 (15.1 stops)
- highlights assumed to saturate in 1/100s at ISO 100
- noise and dynamic range of Canon EOS 1D Mark III simulated
- LDR images saturate at 15280 (may seem underexposed in 16-bit format)
- ground truth -
HDR,
tonemapped
- exposure bracketing ±2 stops -
input,
merged HDR (2.8 dB),
tonemapped
- ISO 100, 3 images (1/100, 1/25, 1/6)s
- our SNR-optimal sequence -
input,
merged HDR (14.6 dB),
tonemapped
- same time budget and #photos as exposure bracketing
- ISO 3200 (1/3200, 1/125, 1/5)s
- SNR graphs - predicted
Real example: Desk still life (Fig. 7)
- HDR photography by Sam Hasinoff
[CC3.0 license]
- dynamic range of 6500 (12.7 stops)
- highlights saturate in 1/800s at ISO 100
- Canon EOS 1D Mark III, 10MP raw images
- densely sampled "ground truth" sequence -
input,
merged HDR,
tonemapped
- ISO 100, 17 images (1/3200, 1/1600, ... , 20)s
- exposure bracketing ±1 stop -
input,
merged HDR (5.9 dB measured),
tonemapped
- ISO 100, 3 images (1/800, 1/400, 1/200)s
- our SNR-optimal sequence -
input,
merged HDR (16.2 dB measured),
tonemapped
- same time budget and #photos as exposure bracketing
- ISO 800 (1/6400)s + ISO 6400 (1/1600, 1/125)s
- SNR graphs - predicted
Acknowledgements
This work was supported in part by an NSERC Postdoctoral Fellowship, NSF CAREER award 0447561, the Quanta T-Party, NGA NEGI-1582-04-0004, MURI Grant N00014-06-1-0734, and
gifts from Microsoft, Google and Adobe.
F. Durand acknowledges a Microsoft Research New Faculty Fellowship and a Sloan Fellowship.
Thanks to Rafał Mantiuk for releasing the HDR Nancy church image
under a Creative Commons license.