aCuria
Offline
Upload & Sell: Off
|
 Re: Re - would you buy 39mpx, 5ms read, 30fps, CFEx | |
ilkka_nissila wrote:
aCuria wrote:
ilkka_nissila wrote:
aCuria wrote:
Scott Stoness wrote:
39mpx APSC R7ii
When you are asking for 39MP on APSC for $2500, you probably presume that such a camera will produce a higher quality image than the R6ii / R6iii ($2700) at 24MP FF
However AFAIK the 24MP FF is going to out-do the 39MP APSC in real life shooting conditions.
Think about it this way, even a "perfect" lens at f/7.1 or darker is unable to produce 45MP of resolution (FF) or 17.5MP (APSC) because diffraction has already set in on such a sensor at f/6.6!
This means, even if you have that 39MP apsc sensor, at f/7.1 you cant record more than 17.5MP of actual resolution regardless of how good your lens is!
There is no sharp boundary beyond which increasing the pixel count produces no improvement in image quality. Diffraction does reduce the MTF at mid to small apertures, but it's still possible to get some improvement by increasing the pixel count. Increasing the pixel count might increase the noise (certainly it does so at the pixel level, but not necessarily at the image level) and this will in practice limit the detail that can be obtained, but in bright light there should still be good detail at f/8 and while f/11 would be less sharp than f/8 it would probably still be a tiny bit sharper on a 39 MP APS-C than with a lower pixel density camera. Personally with 45 MP I've never felt constrained by diffraction and often shoot at f/11 or f/13 in the studio where there is plenty of light and I need to get most of a head reasonably sharp. Yes, these are not the sharpest apertures but it doesn't mean the images fall off a cliff.
This doesn't mean I would be on the market for a 39 MP APS-C camera. I am happy with 45 MP full-frame and if my subject is not defined well enough by that, then I am not close enough or there is not enough light. However, there may be people who do want every bit of detail they can get from their lens, for bird photography and other similar subjects.
While there isn’t a perfectly sharp cutoff in practice, because real images depend on light wavelength, sampling efficiency, and system transfer functions, there is a hard upper bound on resolution set by physics. The Rayleigh criterion defines the maximum resolving power of a lens with a circular aperture at a given f-number. This limit is often expressed in lp/mm, but it can just as well be translated into an equivalent megapixel ceiling.
A better analogy may be:
Imagine capturing an image using the best possible full-frame or larger format sensor under ideal studio conditions, then down-sampling it to 17.5 MP. Lets call this our “ground truth” reference.
Now take an APS-C camera with 39 MP, or hypothetically 100 MP. At f/7.1 or smaller this camera simply cannot resolve more detail than that 17.5 MP “ground truth” reference because of diffraction. Extra pixel count won’t recover information that the optics never delivered in the first place.
While I agree there are practical limitations to resolution determined by diffraction, I believe your analysis is incorrect on the specifics. You need to consider the case of spatial frequencies with different colors (for example luminosity might not change spatially but color might) and the Bayer sensor array has a more sparse set of red and blue filtered photosites, and also consider spatial frequencies in 45 degree diagonal direction. To have no improvement in color detail there should be at least 8*sqrt(2) photosite pitches per spatial frequency cycle (sqrt(2) comes from diagonal direction, a factor of three or four is needed to correctly sample a frequency because there are some phases where the two samples dictated by Nyquist would lead to zero contrast, and a factor of two comes from the Bayer array 2x2 matrix, totalling 8*sqrt(2)).
If you look at test shots made by the digital picture com, e.g. the 135/1.8 on a Sony A7RIV,
https://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=1421&Camera=1442&Sample=0&FLI=0&API=7&LensComp=0&CameraComp=0&SampleComp=0&FLIComp=0&APIComp=0
If you select aperture f/16 you can still see a ton of moire in the center of the image, suggesting the camera has inadequate density of photosites to render the image correctly even at the aperture considerably blurred by diffraction at f/16. The sensor resolution must be increased considerably (beyond the 61 MP FF, equivalent to 23 MP 1.6x crop) before we can expect color correct fine detail to be rendered in our images, assuming a tripod and stationary subject at low ISO. At f/7.1 the issue is obviously all the more pertinent. The 400/2.8 similarly shows moire at f/16 on a 50-megapixel camera (61 MP not shown by this tool). Only when moire completely disappears is the image true to the subject.
In practical shooting situations, low light, noise, and subject movement blur reduces the system resolution and in practice in many cases those gains beyond let's say 45 MP full-frame might not be realized except for static subjects in bright lighting conditions, but it's totally possible that in some practical situations moire does show up. I tend to avoid the sweet spot apertures and shoot either near wide open (f/1.4 to f/2.8) or well stopped down (f/11), to avoid these problems. However, when I was shooting with the Nikon 500 PF which is a very high resolution lens with minimal CA, at f/5.6 I was getting erroneous bird color detail a lot of the time on 45 MP sensors. To get correct rendering of the subject in those circumstanecs, the pixel count must be increased since stopping down would introduce noise (due to practical lighting conditions that I had). Stopping down does effectively help mitigate moire and this is a benefit of diffraction helping out the photographer. When using studio lighting, stopping down can be achieved without introducing noise.
The f/7.1 lens will resolve less than 17.5MP (apsc) or 45MP (FF) of luminance data.
Its true that if the sensor resolution exceeds lens resolution then luminance moire is suppressed.
However this does not mean you wont see color moire. This is caused by the bayer pattern and associated non-uniform color sampling, interpolation and demosaicing.
I do think the next breakthrough in sensors would be a miniaturized 3CCD, or advanced Foveon type sensor. This would solve both color moire, and potentially improve iso sensitivity by at least 1 stop since green light is not thrown away at the red and blue photosites and vice versa.
If you try using 4 way pixel shift, which simulates the foveon sensor, you will find that the color moire is suppressed even though pixel count did not increase.
|
