by Roger N. Clark
Photographers often want to upgrade their gear to have better capability. Often, this capability includes better low light capability. Other reasons to upgrade are faster autofocus performance, faster frame rate, more megapixels, etc. Which will give better low light performance, a camera or lens upgrade?
Exposure and Digital Camera Series:
Prerequisite Before reading this article, please read the other articles in this sequence (links above).
Photographers often want to upgrade their gear to have better capability. Often, this capability includes better low light capability. Other reasons to upgrade are faster autofocus performance, faster frame rate, more megapixels, etc. This article will address only trying for better low light performance.
Before proceeding, be sure to understand what is ISO, the difference between camera exposure and true exposure that where were discussed in the prerequisite articles in the above link.
Example Upgrade Question
I currently have a Canon 7D and a 400 f/5.6 lens and would like to do bird photography in low light conditions like jungles. Given limited funds, should I upgrade to a different camera body or a better lens? If camera body, which one, or if a lens, which one. I'll give answers in general terms, so it need not be specific to these models or manufacturer.
Online suggestions to questions like the above seem to have more votes for upgrading the camera than upgrading the lens. I believe the reason is that the common "wisdom" on the internet is that larger pixels gather more light, so upgrading to a camera with larger pixels is a solution.
From the prerequisite article, Exposure, Light Meters, and Digital Cameras, it should be clear that larger pixels do collect more light, but at the expense of spatial detail, or pixels on the subject. To compensate, people suggest using a longer focal length, such as getting a teleconverter. With the longer focal length, one can restore the pixels on subject that one had with the shorter focal length lens and camera with smaller pixels. The problem with this idea is that nothing changed in basic light gathering from the lens. Adding a teleconverter does not change the amount of light collected, so the total light from the subject remains the same. Adding a teleconverter spreads the light out more, and when used with a camera with larger pixels simply changes the bin size for collecting that fixed amount of light. Light meters will generally change the exposure time when the teleconverter is added, and changing exposure time will change the amount of light collected. But that could have been done with the original camera and lens with no teleconverter. And if one is limited in exposure time (e.g. due to subject movement), then lengthening exposure time is not an option to gather more light.
Let's work an example. The Canon 7D has 4.3 micron (0.0043 mm) pixels, and a 400 mm f/5.6 lens has an aperture diameter of 400/5.6 = 71 mm. At 400 mm, the angular size of a pixel is (1000 * 0.0043/400) = 0.011 milliradian (mr). See the prerequisite articles for more info on these types of calculations.
If we say that the angular size of 0.011 mr is the minimum we want to maintain in a new camera/lens upgrade, what do we need to achieve better light collection? This sets our "subject" size as 0.011 mr, so the Etendue on the subject is:
EtS ∝ A * Ω(subject),
A = lens aperture area (more precisely, the lens entrance pupil area),
A = pi * D2 /4. where D = Lens diameter (entrance pupil diameter),
D = Lens diameter = focal length / f/ratio,
pi = 3.14159, and
Ω (omega) is the solid angle of the SUBJECT.
Note, the ∝ means proportional to.
Thus, for the 400 mm f/5.6 lens, A = 3.14159*712/4 = 3959 square mm, and
EtS ∝ 3959 * 0.0112 = 0.48 (units are mm2 mr2)
Now let's upgrade the camera to a camera with 1.4x larger pixels, or 6.0 microns. But to compensate for the larger pixels and less pixels on subject, we add a 1.4x teleconverter. The lens area is still the same, 3959 mm2. And the angular size of the pixel is (1000 * 0.0060/(1.4*400)) = 0.011 mr.
Well, guess what, the Etendue factor, EtS, of the new combination is:
EtS ∝ 3959 * 0.0112 = 0.48 (units are mm2 mr2),
which is exactly the same as before. We have not improved low light capability. But as a side effect, we increased the f/ratio of the lens to f/8, which slows down the AF system and reduces AF accuracy in low light. The result of this camera upgrade is no change in low light performance, but a degradation in autofocus performance.
To maintain subject resolution (pixels on subject) for the same exposure time, AND improve low light capability, the only solution is to increase the lens collecting area. In the above equations, if we keep the pixels on the subject the same (0.011 mr angular pixel size), the only other variable in the equation is lens area. Thus lens area is the key to better low light capability (this assumes the different cameras have the same quantum efficiencies, fill factors, and filter transmissions, which is pretty much the case for digital cameras produced in the last few years).
If you read the prerequisite articles, you should see a consistent theme here: the lens is the key to light gathering, NOT the pixel and the sensor. The pixel and sensor are simply carriers of the photelectrons; it is the lens that collects the photons.
Thus, for this question, it is basic physics that for better low light performance, upgrade the lens. From a 400 f/5.6, the next available lenses would be a 400 f/4, or 300 f/2.8 with a 1.4x TC (giving 420 mm at f/4). The 400 f/4 has an aperture diameter of 100 mm, while the 300 f/2.8 has a 107 mm diameter lens. Thus the 300 f/2.8 + 1.4x teleconverter would collect slightly more light from the subject than the 400 f/4 lens. Further, the autofocus system would be working at f/4 compared to the camera only upgrade working at f/8.
Upgrading from a Camera with Small Pixels to one with Larger Pixels for "Better Noise"
Consider the following. A bird photographer has a Canon 7D and a 400 mm f/5.6 lens. The photographer is considering changing to a Nikon D700 for better low light/high ISO capabilities. This is a common idea on the internet these days. Will changing to a camera with larger pixels really help?
The 7D has 4.3 micron pixels while the D700 has 8.45 micron pixels. This translates to pixel areas of 18.5 square mm for the 7D and 71.4 square mm for an area ratio (D700 pixel area/7D pixel area) of 3.86, or about 2 stops.
So, for the same sensor sensitivity (which will be very close, probably within 10 percent or so), a D700 will collect almost 2 stops more light per pixel from the same lens and same exposure time and f/ratio (note, this is independent of ISO; remember ISO does not change sensitivity).
But the D700, with its larger pixels, has less pixels on the subject, so the image has less resolution on the subject. Linearly, this would be 4.3/8.45 = 0.51x fewer pixels with the D700. Area would be 0.26x fewer pixels, or exactly the same fewer pixels by area as the per pixel area ratio, and the reduction in light per pixel for the same camera exposure. It is simple math to sum up the light per pixel to find the total light from the SUBJECT is the same (7D: 1/4 the light per pixel but 4 times more pixels).
The bottom line is pixel size directly trades resolution (pixels on subject) for more light per pixel. This is what we perceive as lower noise in the images from the camera with larger pixels.
But if one's goal is to have the same pixels on subject with the D700 as with then 7D, the solution (which will be close, within about 4%) is to add a 2X TC on the lens when on the D700. But what do you do regarding exposure when you add a 2X TC? Either 1) increase ISO two stops, or 2) increase exposure time by two stops, or 3) open the aperture by 2 stops (if possible). If #1, we get a noisier image (because ISO does not increase sensitivity), and we did not change the true exposure. #2 and #3 increases exposure so improves the signal-to-noise ratio.
Now, the 7D with the same lens as used above gets the pixels on the subject that the d700 needed a 2X TC to achieve. Functionally, the smaller pixels of the 7D are the same angular size as the pixels for the D700 using a 2x TC on the same lens. Same pixels on subject. Same depth of field. Same light per pixel. This is what Etendue is: the Etendue of the two cameras are equal. But by the way the camera manufacturers have defined camera exposure, the meter says use the same exposure on the subject that the D700 with no TC says to use on the subject. But we see this results in less exposure per pixel and more noise with the 7D.
To expose the 7D image like that with D700 and the 2x TC, simply treat the 7D like is has a built in 2x TC (compared to the D700 pixels), and thus increase exposure by two stops (longer exposure time or larger (faster) aperture). This then equalizes the Etendue, the pixels on subject, the light per pixel and the exposure. Then the noise will appear the same in the images from both cameras.
So try this to prove the concept: Find a static scene and image it with the 7D. Next use the same lens on the D700 (assuming you have the same focal length and f/ratio lenses for the two cameras). and set the f/ratio the same as you used on the 7D (let's say it is f/4). Now add a 2X TC (the f/ratio will now change 2 stops, e.g. to f/8) and expose manually with the same exposure time as on the 7D. The signal-to-noise ratio will be the same in the 7D and D700 images.
There is yet another way to approach the noise difference between large and small pixels. Above we saw that adding a 2x TC to the D700 camera brought the pixels on subject to the same amount as in the D700 images. But another way to reduce the apparent noise in the 7D images is to average 2x2 pixels. This brings the pixel size to almost exactly the same as that for the D700. Then the apparent noise between the two cameras will be the same.
In the Etendue explanation, the A*omega product is equalized. The A is the lens area and the omega is the angular area of the pixel. One can manage the angular area of the pixel by changing focal length, or changing pixel size. If changing pixel size, then there is no difference in signal-to-noise ratio, whether using a different camera with different sized pixels, or one post processes and adds pixels together. By using the same lens, the A (lens area) is the same, and we change omega by changing cameras, focal length or adding pixels together. (Photoshop is behind the times and doesn't add pixels). There is another side effect of adding pixels together, and that is the resulting image has sharper pixel to pixel contrast (this was shown in part 2 of this series: Exposure and Digital Cameras, Part 2.
So adding pixels together with the 7D in 2x2 averages not only produces apparent noise equal to that from the D700, the 7D image will appear slightly sharper. This, of course, assumes the subject fits in the frame of the 7D and the same aperture diameter lens and same focal length is used on both cameras.
Figure 1 is another example using a 7D at ISO 6400, which produces a quite noisy image. Compare that to the 5DII, with its larger pixels produces an image with less apparent noise, but lower spatial resolution, and then compare to the 7D with pixels scaled then averaged 2x2: a pretty clean image in comparison. The scaling up by 1.35x does not change the noise, and the 2x2 pixel average simply has the omega in the A*omega equation made larger and equal to that for the 5D Mark II image. Note that the TOTAL light from the subject is the same for the original 7D image and the scaled, 2x2-pixel averaged image. The noise, however, appears less in the resized 7D image than that in the 5D Mark II image. The is because the 7D sensor has slightly greater system sensitivity than the older generation 5D Mark II, so the 7D collected a little more light, making a higher signal-to-noise ratio image. This proves the 7D is actually a better low light camera than a 5D Mark II.
Figure 1. Comparison of 7D, 5DII and scaled 2x2 pixel averages of the 7D image at ISO 6400. The 2x2 pixel average was performed in ImagesPlus. This rescaling and pixel averaging make the image the same size as the Moon in the 5D Mark II image. If the two cameras had equal system sensitivities, the noise would be identical. The rescaled 7D image on the right actually shows less noise than in the 5D Mark II image because the 7D has greater system sensitivity. No sharpening and no noise reduction software was run on any image.
Again Etendue explains all the situations. Understand Etendue and manage noise, resolution and exposure better and how they trade in real world imaging.
The answer, to the question will upgrading to a camera with larger pixels help reduce apparent noise in images if using the same lenses is two fold. Yes, the apparent noise will be less but at the expense of less resolution (pixels on subject). If one wants to maintain pixels on the subject, then one would need to increase focal length by adding TCs, and for the same exposure time, there would be no difference in noise between the large and small pixel cameras.
The lens is key to collecting light. Always choose the lens with the largest diameter aperture to collect to most light. This is independent of f/ratio, as we saw in part 2 of this series.
The caveats to this rule regards sensor efficiency. If you have a very old digital camera, e.g. made before circa 2009, a newer cameras with a more modern sensor will help a small amount, but usually not as much as a lens upgrade. Cameras may improve in efficiency further, but all tests I have conducted on digital cameras circa 2009 - 2013 have similar sensitivities. (See Digital Camera Sensor Performance Summary for more information. See Figures 4a and 4b at: Nightscape Photography with Digital Cameras for a comparison of Canon 7D images versus cameras with larger pixels, proving that low light sensitivity does not improve with increasing pixel size.
The above example also applies to wide angle lenses and things like nightscape photography. If the subject fits in your field of view, choose the lens with the largest clear aperture to collect the most light. Lens clear aperture is the key to light collection from the subject, not f/ratio, and not pixel size. Pixel size on the subject can be managed in post processing, trading less spatial resolution for less apparent noise.
The bottom line is understanding Etendue enables one to manage true exposure on the subject, as well as camera exposure to achieve the noise one wants in one's images. If you have cameras with different sized pixels, and if you want the same noise appearance, treat the cameras with smaller pixels as if they had TCs.
Exposure and Digital Camera Series:
First Published May 22, 2013
Last updated October 5, 2014.