Yesterday, I made the following remark apropos of the Sony 100 megapixel sensor:
If your subject will hold still, you could probably duplicate the results with a three-shot pano on the a7RII. Except for the “P1 color” (just kidding about the color, although I bet the CFA is different).
I was challenged by a reader who said that you could achieve the same resolution that way, but not the same signal-to-noise ratio or dynamic range.
This post is about why you can achieve the same SNR and DR.
The pixel pitch of the new medium format Sony sensor, at 4.6 micrometers, is essentially the same as the current sensor in the Sony alpha 7RII. Let’s assume that the design of the sensor pixels in the two cameras is similar.
This is a little risky, since
- Phase One has said nothing about change in conversion gain with ISO setting, which is a trick that the a7RII uses to reduce noise, increase signal-to-noise ratio, and increased dynamic range that ISO settings of 640 and above. If the new Sony 100 megapixel sensor doesn’t have similar technology, the a7RII will have an advantage at ISO settings about 640.
- Also, Phase One is silent at this point on whether the new sensor is backside eliminated (BSI) or front side illuminated. The a7RII is a BSI sensor, which should give it somewhat of an advantage and the SNR and dynamic range department. As an aside, BSI would help with corner color casts, which bedevil the MF high-pixel-count sensors with some wide lenses.
- On the other hand, the new 100 megapixel sensor is eight months younger, and therefore the technology it employees might be somewhat better, especially since its base ISO is 50, not 100 like the a7RII.
Without inside knowledge of any of the above characteristics of the new chip, I’m going to plow ahead assuming they’re the same at the pixel level.
I’m also going to assume that the CFA doesn’t affect the DR or SNR. This is probably not strictly the case, since MF cameras have traditionally used CFAs with more spectral overlap, improving capture metameric error at the expense of SNR.
The fact that the lens used on the a7RII for the pano doesn’t have to cover the final stitched format makes it easier to design it, and probably favors its flare performance, which will affect captured dynamic range. We’ll ignore that, too.
So, at the pixel level, if the pixel designs are the same on the two sensors, the read noise, the conversion gain, and the full-well capacity are all the same. Let’s assume, as it is certainly true in the 14 bit (at best) a7RII, that the read noise is sufficient to properly dither the analog-to-digital converter, and thus that the optional 16 bit precision of the new sensor offers no advantage in reducing read or photon noise. So the pixel level engineering dynamic range and the pixel level photon noise of the two cameras must be the same.
When photographic dynamic range, as opposed to engineering dynamic range, is quoted, the images are almost always normalized to the same size.
So this talk about how stitching affects image size at the same angle of view.
Let’s say we take a picture with a 100 mm lens set at f/5.6 with the new medium format camera. The dimensions of that picture will be 11608 x 8708 pixels. If we put the a7RII in portrait orientation, and do a three-shot pano with the same lens sent to the same aperture, will get an image whose horizontal dimension is limited by overlap, and whose vertical dimension is 7952 pixels. If we raise the overlap so that the pixel count of the two images is the same, the stitched image will have an orthogonal dimension of 12711 pixels, and a diagonal of the two images will also be the same.
If the diagonal pixel count is used for normalization to photographic dynamic range, the PDR of the two images from the two cameras must be the same, to the characteristics of each pixel are the same and when normalizing to the same diagonal.
It’s even easier to see what is going on if we do the following: experiment. Imagine taking the same 100 mm medium format lens as in the above example and take a single shot with the 100 megapixel sensor. Now put that lens on the tail/shift adapter as slap the combination of the a7RII in portrait orientation. Set the tilt to neutral. Make three exposures, the first with the shift set to neutral, the second with the body shifted right, and the third with the body shifted left, and combine the three images into a panorama. You can see that we have achieved the same effect that we would have gotten with the larger sensor by assembling pieces of the larger image from the smaller sensor.
Many people don’t realize that SNR is dependent on the detail level you consider, and thus photographic dynamic range is dependent on the magnification from sensor size to display size.
Yes, I’ve covered that before, but it rapidly gets complicated, which is why I didn’t go into any details here. That kind of thinking is the basis of the Claff PDR, which I use.
At 50,000 feet, halving the resolution combines four pixels into one, which doubles the SNR, since the signal adds algebraically, and the noise adds in quadrature, thus, four times the signal and twice the noise. Lotta assumptions in there, though.