Canon 70D long exposure sensor noise comparison

I tend to lurk around the forums at cloudynights.com.  Recently, I have been looking into getting a new camera for doing astrophotography.  This was partly prompted by some nice, short exposure photos at ISO 6400 that my brother took with his Nikon D600.  My micro four thirds Panasonic DMC-G3 just cannot compete with his full frame sensor.  "Amp glow" in the DMC-G3 makes the top 1/6th of the sensor close to useless for any exposures over 30 seconds, and forget about using ISO settings beyond 400 due to noise.

For my (microprocessor controlled) astrophotography goals, and for my budget, I need camera with little noise at high ISO values and long exposures.  A "soft" requirement is an articulated LCD screen so that I don't have to kneel on the ground to see what is being imaged through my refractor (yes, I could raise the tripod legs at the risk of decreased stability).  DXOMark.com does comparisons of many cameras, so I looked at the information they had.  Most of their testing is geared towards daylight photography, but they do some tests for "Sports (Low-Light ISO)" which is a starting place for us.

Make	Model	Low-Light ISO Score
Panasonic	DMC-G3	667
Panasonic	DMC-GH3	812
Canon	60D	813
Nikon	D5000	868
Panasonic	E-P5	895
Canon	70D	926
Nikon	D5100	1183
Nikon	D5300	1338
Nikon	D610	2925

A full frame sensor (which scores great) is not in my budget, but I added the Nikon D610 to the list for a comparison.

So, in my lurking, I came upon the possibility that a number of cameras are amenable to modifications that allow for H-alpha wavelengths to be captured by CMOS sensors in digital cameras.  To be more accurate, the CMOS sensors are already able to capture H-alpha wavelengths (and IR and UV).  Camera manufacturers put glass and coatings over the sensors to block out UV and IR light so that terrestrial photography represents what our eyes see.  Canon has two models, the 20Da and 60Da that come from the factory with the modification.  Third party services such as Hypercams & Mods do the modification for reasonable fees, and there is always the DIY approach.

For the price and DXOMark performance, the Nikon D5300 is looking great against similar competitors, but there is some concern about Amp glow.  With more lurking, I came across a nice comparison of ISO and exposure times for the Nikon 5300.  Being a scientist, I thought that there might be a systematic way we can evaluate the performance of digital camera sensors for astrophotography (and in general for long exposures).  I saw a Canon 70D on sale, so I picked it up for testing (since I could quickly obtain a Canon EF to T-ring adapter for astrophotography).

Materials
Canon 70D
Michron intervalometer and E3 cable
iPhone 5

Methods
I configured the Micron to capture the following series of images with delays:
24 images x (1 minute exposure with 15 seconds delay)
20 images x (2 minute exposure with 15 second delay)
21 images x (4 minute exposure with 15 second delay)
15 images x (8 minute exposure with 15 second delay)

The delay was chosen to allow the sensor to cool down a bit between image captures.  The camera and Micron with cable were placed in my basement with lights out.  The temperature was set to 69° F.  Image recording was set to obtain JPG + RAW images at full resolution.  Noise reduction in the camera was turned off for long exposures.  The lens of the camera was removed and replaced with the body cap.  The articulated screen of the 70D was opened to provide better ventilation.  An ISO setting was chosen, and the intervalometer was started.

By taking multiple photos using the intervalometer function of the Michron, the camera's sensor was allowed to warm up to a rough equilibrium point.  The final JPEG image of each exposure time series was then used for image comparisons.

As a control the same procedure was done on a Panasonic DMC-G3.  "Bulb" exposures are limited to 2 minutes on the G3, so data could not be collected at 4 and 8 minute exposure lengths.

Results

Figure 1:  Canon 70D. Final JPG image of each exposure series was scaled to 5% to show overall sensor noise profile.

Figure 2: Canon 70D. Final JPG image of each exposure series was cropped to 274x182 pixels, center weighted, to show detailed noise and "hot" pixel profile.

Figure 3: Control, Panasonic DMC-G3.  Final JPG image of each exposure series was scaled 5.3% to show overall sensor noise profile or cropped to 244x182 pixels, center weighted, to show detailed noise and "hot" pixel profile.

From a visual comparison of the center cropped images, the Panasonic DMC-G3 at 1 minute exposure performs roughly equivalent to the Canon 70D at 8 minutes exposure. While the 70D is uniformly dark at ISO 200 and 400, the G3 has noticeable gray 2x4 pixel marks over the entire sensor.

The 70D is relatively free of banding, but there is some vertical banding observed at high ISO and long exposures.  The DMC-G3 has a noticeable band across the top of the image, plus a diagonal fault of some sort.

Figure 4: A comparison of RAW file size versus exposure time for selected ISO values

The recorded value of each pixel, in terms of data, will depend on the amount of light (or noise) captured.  Consequentially, the final RAW image size on disk will correlate with the amount of noise captured at the given settings.  By plotting image size versus exposure time (Figure 4), we can see that a doubling of exposure time roughly correlates with a halving of ISO in terms of sensor noise.  Additionally, at higher ISO settings, a doubling of exposure time shows a much greater increase in sensor noise than at lower ISO values.

Discussion
For astrophotography and general long exposure purposes, the tests represent a best case scenario.  Light pollution will reduce the reasonable exposure/ISO values, but this can be partly offset by a light pollution filter.  What we can take away from these test are the following.  First, the 70D sensor readout is fairly uniform.  There does not seem to be any unusual "amp glow", and "banding" is minimal.  Second, we can subjectively observe when "hot" pixels start to be a nuisance.  An objective test would be to determine when the first hot pixel reaches saturation, since dark frame subtraction would leave a hole.

From experience, the DMC-G3's gray 2x4 pixel marks wind up making long streaks on stacked images if there is any drift in your telescope tracking, even with dark frame subtraction.  This effectively limits the DMC-G3 to ISO 200 at less than 2 minute exposures or ISO 400 at less than 1 minute exposures.  The 70D does not seem to have similar, inherent noise issues.  However, the sensor noise in the 70D seems to be more "blobby" as compared to the granular noise in the DMC-G3.

Sensor noise increases significantly more at higher ISO values as the exposure time increases.  If we declare an acceptable amount of noise, for instance at ISO 800 and 4 minute exposure with the 70D, our plot of file size versus exposure time gives alternate ISO/exposure combinations which are roughly equivalent in terms of noise.  Therefore, if our mount tracking is fantastic, an 8 minute exposure at ISO 200 might help us get those rare photons from a deep sky object.  With a less precise mount (or no mount at all) a 1 minute ISO 800 exposure would reduce the degree of start trails and have roughly equivalent, and actually slightly less, sensor noise.