ClarkVision.com

The Signal-to-Noise of Digital Camera images
and Comparison to Film

by Roger N. Clark

People often say that digital camera images have much lower noise, and this is true. But how much better? People wait for the next generation of cameras to come out with lower noise, but are we reaching possible limits? On this page I present some measurements that help quantify where the technology is now and show at least some cameras are at the limits in some respects.

All images, text and data on this site are copyrighted.
They may not be used except by written permission from Roger N. Clark.
All rights reserved.

If you find the information on this site useful, please support Clarkvision and make a donation (link below).

Test were conducted to characterize the noise as a function of signal level for a Canon 1D Mark II digital camera. A series of different exposures, less than about 1/1,000 second, on a gray scale test target were made with the camera lens out of focus to reduce target non-uniformities. Each level was analyzed for signal and noise as a function of multiple exposures at each ISO setting, and at multiple ISO settings as a function of intensity of the scene. Image date were calibrated linearly to scene intensity for both the digital sensors and film. Similar work was done on scanned slides of Fujichrome Velvia, isO 50 film. The standard deviation at full image resolution was computed using ImagesPlus over a 29x29 pixel box. Multiple areas were checked for consistency in the derived standard deviation. The signal-to-noise ratio is the linear signal calibrated from the linear output of a digital sensor divided by the standard deviation at that intensity. The intensity levels were confirmed by meter readings. Scanner noise was checked and had a negligible effect on the results. The results are shown in Figure 1, below.

Figure 1. The signal-to-noise of the Canon 1D Mark II camera at each of its ISO settings is compared to film and the sensor's theoretical maximum. The theoretical curve assumes the signal-to-noise is limited by photon statistics and the full well capacity of the sensor (see Table 3) and not by noise in electronics in the camera. This theoretical maximum curve is independent of the quantum efficiency of the sensor, and independent of the transmission of the optics. Also compared is the signal-to-noise derived from the fine-grained film, Fujichrome Velvia (ISO 50). The data in these plots are 16-bit unsigned integer values, scaled by a factor of 16 from the camera DN (see notes at the end of this page) when the camera data were converted from the native raw format.

In Figure 1, we see the signal-to-noise of the Canon 1D Mark II at ISO 100 is compared to the sensor's theoretical maximum. This maximum curve was computed assuming all the noise is due to photon statistics and the derived full well capacity (at Camera DN 4095) from Table 3. Within the accuracy of the data, it shows that the 1D Mark II at ISO 100 is essentially photon noise limited. This is indicating that these high-end cameras are reaching the limits of what is possible from current technology.

The Canon 1D Mark II results mean that the only way to make a sensor with higher signal-to-noise using this same technology is to produce one with a greater full-well capacity and then somehow deliver more photons to the sensor. The reason for this is the theoretical maximum signal-to-noise is controlled by how many photons you count. sending more photons to the sensor (e.g. increasing the light intensity, getting a faster f/stop lens) would not improve the maximum signal-to-noise, unless those additional photons could be counted so you would need a large well to collect them.

Ways to improve upon the delivery of photons to the sensor includes the following.

1) Improve the quantum efficiency of the photo sensitive device. Current quantum efficiencies of CCDs and CMOS sensors is in the 20 to 50% range (depending on wavelength).
2) Improve the transmission of the colored filters and blur filter over each sensor.
3) Improve the transmission of the lens on the camera.

A theoretically perfect sensor would have 100% quantum efficiency, and 100% transmission of the optics. It doesn't exist. But you would still need a larger well to improve the maximum possible signal-to-noise. Improving transmission and quantum efficiency would, however, mean you coudl reach the maximum signal to noise at a higher ISO. For example, doubling the quantum efficiency of the 1D Mark II sensor with no other changes would mean the sensor theoretical maximum curve in Figure 1 would occur ar ISO 200, not the current ISO 100.

The comparison in Figure 1 shows that the Canon 1D Mark II has a signal-to-noise at ISO 800 that is still better than fine grained ISO 50 film. At ISO 100, the 1D Mark II has signal-to-noise on the order of 3 to 6 times higher than Fujichrome Velvia, ISO 50 slide film. Higher speed films have greater grain and lower signal-to-noise.

Figure 2. The signal-to-noise of the Canon 1D Mark II camera at each of its ISO settings (points and lines in color) is compared to the maximum theoretically possible from photon (Poisson) statistics (black lines). The measured values (points and lines in color) show that within measurement error, the 1D Mark II is performing at the theoretical maximum.

The theoretical curves in Figure 2 use the measured full well capacity at ISO 100 of 52,000 electrons. The maximum signal-to-noise is then (52000/(square root 52000) = 228. At higher ISO values, no additional photons are collected, only the gain of electrons in the well to output DN is changed. So at ISO 200, the maximum signal is half of 52,000 or 26,000 electrons and the maximum signal-to-noise drops to (26000/square root 26000) = 161. Each increase in ISO of a factor of 2 drops the signal-to-noise by square root 2, so at ISO 1600, maximum signal occurs at 3250 electrons and a maximum signal-to-noise of 57 is all that can be achieved.

Below are tables that give other derived parameters for the 1D Mark II camera along with data from the internet for comparison. Methods for determining gain and read noise can be found at references 1-5.

Table 1

                 Gain
             (electrons/DN)
        -----------------------------------------------------
         1DMII   5D    20D   10D   350d  300D    D70    S60
  -----------------------------------------------------------
ISO  50   26.03 32.6                                    5.4
ISO 100:  13.02 16.3  12.4  11.4  10.2   11.1           2.7
ISO 200:   6.51  8.2   6.2   5.5   5.1    5.6    6.0    1.3
ISO 400:   3.25  4.08  3.1   2.7   2.56   2.78   2.98   0.7
ISO 800:   1.63  2.0   1.5   1.4   1.3    1.4    1.34
ISO1600:   0.81  1.0   0.8   0.7   0.6    0.7
ISO3200:   0.41  0.5   0.4

Canon 1D Mark II values from this study.
Canon 1DMII are newer values determined Feb 12, 2006 with firmware 1.2.4.
Canon 10D values from Tam Kam-Fai posted on digital_astro@yahoogroups.com, 20D, 300D, D70 ISO 400 values from Terry Lovejoy ( http://www.pbase.com/terrylovejoy/image/30685101 ) Canon S60 5-megapixel point and shoot digital camera from this study.
The 5D, 350D ISO 400 values are from reference 13. Other values compured from those at ISO 400. The 20D value also agrees with reference 13 where 3.09 electrons/DN is reported. Reference 13 reports the 10D at ISO 400 has a gain of 2.34 electrons/DN, 15% lower than above.

Table 2

               Read Noise
              (electrons)
        ----------------------------------------------------
         1DMII   5D    20D   10D   350d  300D    D70    S60
  ----------------------------------------------------------
ISO  50:  30.6  59.7                                   13.6
ISO 100:  16.6  30.1  25.3  15.9   21.6
ISO 200:   8.95 15.6  13.5  11.0   11.5
ISO 400:   5.56  8.4   7.5  10.6    7.2           6.3  
ISO 800:   4.04  5.2   4.8   9.0    4.9    10    13
ISO1600:   3.90  3.7   3.6   9.0    3.7
ISO3200:   3.93  3.7

Full well depth (electrons for max DN at iso 100) (maybe we should call this the "camera maximum DN well depth", because it is not necessarily the real full well depth). Canon 1D Mark II values from this study. Canon 1DMII values determined Feb 12, 2006 with firmware 1.2.4.
Canon 10D values from Tam Kam-Fai posted on digital_astro@yahoogroups.com, 20D, 300D, D70 values from Terry Lovejoy ( http://www.pbase.com/terrylovejoy/image/30685101 ) and http://www.astrosurf.org/buil/20d/20dvs10d.htm The Canon 5D, 350D, and 20D values computed from the gains above and read nouse in DNs from Table 3 of Reference 13.

Table 3

                   Sensor     
                full   Read   Sensor Dynamic range   Pixel
Camera          well   Noise  (full well/read noise) Spacing              Sensor size
                 (electrons)   linear*       stops  (microns) Mpxl    pixels         mm
Nikon D2Hs        ?                                    9.4     4.0 2464 x 1632   23.1 x 15.1
KAF-18000CE   100,000   18      5560         12.4      9.0    18.0 4904 x 3678   46.05x 35.0
Sigma SD10        ?                                    9.0     3.5 2304 x 1536   20.7 x 13.8
Canon 5D      ~80,000?   3.7  ~20000?       ~14.3?     8.2    12.7 4368 x 2912   35.8 x 23.9
Canon 1DMII    79,900*   3.9   20500         14.3      8.2     8.2 3504 x 2336   28.7 x 19.1
Nikon D70      48,800    6.3    7740         12.9      7.9     6.0 3008 x 2000   23.7 x 15.6
Pentax *ist Ds    ?                                    7.8     6.0 3008 x 2008   23.5 x 15.7
Canon 10D      44,200   10      4420         12.1      7.4     6.3 3072 x 2048   22.7 x 15.1
Canon 300D     45,500   10      4550         12.1      7.4     6.3 3072 x 2048   22.7 x 15.1
Canon 1DsMII      ?                                    7.2    16.6 4992 x 3328   36   x 24
Canon 20D      51,400    7.5    6850         12.7      6.4     8.2 3504 x 2336   22.5 x 15.0
Canon 20Da     51,400    3     17130         14.1      6.4     8.2 3504 x 2336   22.5 x 15.0
Canon 350D        ?                                    6.4     8.0 3456 x 2304   22.2 x 14.8
Nikon D200     32,680    7.4    4416         12.1      6.1    10.0 3872 x 2592   23.6 x 15.8
Nikon D2X         ?                                    5.5    12.2 4288 x 2848   23.7 x 15.7
Olympus E300      ?                                    5.3     8.0 3264 x 2448   17.3 x 13.0

Canon S60      22,000   13.6    1616         10.7      2.8     5.0 2592 x 1944   7.18 x 5.32
Nikon 8800                                             2.7     8.0 3264 x 2448   8.80 x 6.60
Canon S70       8,200    3.2    2562         11.3      2.3     7.1 3072 x 2304   7.18 x 5.32

Notes:
Inferred full well depth (at max DN of 4095) = gain * 4095 * iso (at that gain) / full_well_iso, where full_well_iso is the lowest iso where the camera reaches full well (=100 for DSLRs and 50 for the S60 and S70 point and shoot cameras). Example, the Canon 1D Mark II at iso 800 has a gain of 1.6, so: 1.6 * 4095 * 800/100 ~ 52400.
*At ISO 100, the Canon 1D MII records a maximum of 52,300 electrons; at ISO 50, 79,900 electrons are recorded, but that occurs about 3/4 of the 12-bit linear scale, at 3071 on the 12-bit DN ramge.
The Kodak KAF-18000CE is targeted as a medium format sensor; see reference 12.
Possible dynamic range of the sensor is theoretical and in practice is limited by the 12-bit (or 10-bit) analog to digital converters in these cameras.
Sensor sizes from manufacturer's data sheets or product reviews.

Full Well Capacities of some digital cameras/sensors:

Table 4

camera or  Ref  mega     full   read
sensor      #   pixels   well   noise   comments
KAI-11000CM 3     11    50,000   34
YM-3170A    4    3.17   35,000   20
Nikon
  DXM1200F  6    1.5    18,000        6.45 micron pixels
KAF-0402ME 10                    21   gain = 1.97 e/DN, Audine camera

Table 5

                                  ISO 100
               Maximum    Possible Signal-to-noise Pixel
Camera          Signal     Maximum    18% Gray    Spacing         Sensor size
             (electrons)                card     (microns)    pixels           mm
---DSLRs ---
Canon 1DMII    52,300        229         97         8.2    3504 x 2336     28.7 x 19.1
Canon 10D      44,200        210         89         7.4    3072 x 2048     22.7 x 15.1
Canon 300D     45,500        213         90         7.4    3072 x 2048     22.7 x 15.1
Nikon D70      48,800        221         94         7.9    3008 x 2000     23.7 x 15.6
Canon 20D      51,400        227         96         6.4    3504 x 2336     22.5 x 15.0
---P&S ----
Canon S60*     22,000        105         44         2.8    2592 x 1944     7.18 x 5.32
Nikon 8800                                          2.7    3264 x 2448     8.80 x 6.60
Canon S70                                           2.3    3072 x 2304     7.18 x 5.32
Panasonic
  Lumix FZ20                                        2.2    2560 x 1920     5.76 x 4.29

Signal-to-noise assumes photon noise limited. Read noise, and other factors can only degrade this number (read noise is insignificant for the maximum possible and 18% gray card signal-to-noise ratios for the cases shown here). * The Canon S60 full well is for ISO 50. P&S means point and shoot. The Canon 20D full well and signal-to-noise is conditional on an initial number that may have a large error bar.

If you find the information on this site useful, please support Clarkvision and make a donation (link below).

References

1) http://www.pbase.com/terrylovejoy/image/30685101

2) http://www.axres.com/technote1.html .

3) Photography with an 11-megapixel, 35-mm format CCD, Kodak, 2003.

4) http://www.dpreview.com/news/0009/00090603ymedia3mpcmos.asp

5) CCD Chip Data Tables

6) http://www.microscopedealer.com/products/photosystems/nikon_dxm1200f.htm

7) Example data sheet: Kodak KAF - 0261E 512(H) x 512(V) Pixel Full-Frame CCD Image Sensor Performance Specification http://www.datasheetcache.com/Kodak/KAF0261E%20Rev.%20%20D%20Preliminary.pdf

8) CCD Gain. http://spiff.rit.edu/classes/phys559/lectures/gain/gain.html

9) Charge coupled CMOS and hybrid detector arrays
http://huhepl.harvard.edu/~LSST/general/Janesick_paper_2003.pdf

10) Canon EOS 20D vs Canon EOS 10D and Canon 10D / Canon 20D / Nikon D70 / Audine comparison
http://www.astrosurf.org/buil/20d/20dvs10d.htm