Pixel Pooper
Senior Member
Not quite. On a 24" HD monitor, an A7R2 image at 100% equals a 7.4 x 4.9 foot print, or 3.7 x 2.5 foot on a 24" 4K monitor.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Why is DR smaller when FF cameras are used in cropped mode?
- Thread starter Architeuthis
- Start date
m100
Senior Member
- Messages
- 2,744
- Solutions
- 3
- Reaction score
- 1,600
Or the closer you get to a noisy photo the noisier it will look ?
Architeuthis
Leading Member
- Messages
- 653
- Reaction score
- 333
Let me thank all participants for the vivid discussion, I really enjoy and understand better now...
My mistake was to not understand the term PDR exactly. I confused PDR (intuitively) with DR of a single pixel. This DR of a single pixel (is there a technical term for it?) I was trying to read out from Bills graphs. I see now that the sensor size has to be taken into account as an additional factor when one wants to compare DR of the single pixels between different sensor sizes...
When one compares PDR between a MFT (Oly EM1II) and a FF camera (Sony A7R3) the outcome is that at base ISO, PDR is almost 2 EV better with the larger Sensor: https://www.photonstophotos.net/Charts/PDR.htm#Olympus OM-D E-M1 Mark II,Sony ILCE-7RM3
Since the Sony sensor is 384% the size of the Oly, a 2 EV difference is just what would be approx. expected by the sheer size difference, when the single pixels have the same DR...
But how can the single pixel DR of the Oly (with much smaller pixels and lower FWC) be identical (if not slightly better, as implied by the PDR graph) to the Sony with larger pixels (and BSI)?
Wolfgang
My mistake was to not understand the term PDR exactly. I confused PDR (intuitively) with DR of a single pixel. This DR of a single pixel (is there a technical term for it?) I was trying to read out from Bills graphs. I see now that the sensor size has to be taken into account as an additional factor when one wants to compare DR of the single pixels between different sensor sizes...
When one compares PDR between a MFT (Oly EM1II) and a FF camera (Sony A7R3) the outcome is that at base ISO, PDR is almost 2 EV better with the larger Sensor: https://www.photonstophotos.net/Charts/PDR.htm#Olympus OM-D E-M1 Mark II,Sony ILCE-7RM3
Since the Sony sensor is 384% the size of the Oly, a 2 EV difference is just what would be approx. expected by the sheer size difference, when the single pixels have the same DR...
But how can the single pixel DR of the Oly (with much smaller pixels and lower FWC) be identical (if not slightly better, as implied by the PDR graph) to the Sony with larger pixels (and BSI)?
Wolfgang
Think of it this way: I need a 9MP photo:
- Crop: To get proper exposure for the noise floor I want, I can do ISO1600, but my highlights are blown.
- Full sensor: I can underexpose down to ISO400 and add 4 adjacent pixels to make a "superpixel" where I have 4x as much headway.
That's a good example. Without dithering, I have black-and-white (1 bit dynamic range). Dithering expands the dynamic range at the cost of resolution. With classic algorithms (dots), I had a trade-off, where a large dot size gave low resolution / high DR, and vice-versa. With modern ones, the further back I stand, the more area I average over, and the more dynamic range I have, and the less resolution.
It's less technical than sigma-deltas but makes the same point. Thank you!
Nice images of dither patterns. You can step back, to see a small, high-DR image, or step in to see a large, low-DR image.
Last edited:
So why isn't it darker ?
The total amount of light required for a given exposure is directly proportional to the area of the sensor. So the variable "total light" tells us nothing more than the variable "area".
That's more pixels, not more light. Increasing the number of measurements reduces the error. Of course you need a certain amount of light to make each measurement, but in the case of cropping, where the measuring instruments (pixels) are all the same, it's the number of data points that matters.
I don't think that is what we see at all.
Don Cox
I agree. This thread is about cropping from one camera, not about the multiple effects of using different cameras with very different sensors and lenses. So these doll shots are irrelevant.
Don Cox
Now view them both at 100% -- or any other magnification that shows the noise clearly.
In the original post, one image was shown at a higher magnification than the other. This makes the high spatial frequency noise more visible. So does viewing a print from a closer distance.
No. I will not consider that, because this thread is not about comparing cameras. It's about using different lenses (or zoom settings) on one camera.
Comparing cameras involves far too many variables, many of them proprietary information, to be practical.
Don Cox
Nyquist.
The lesser number of pixels in a crop reduces the maximum spatial frequency in line-pairs per image height.
More pixels -- more detail.
More pixels -- higher frequency noise.
If you double the number of pixels per image height, you add an octave of detail and an octave of noise. But you have to enlarge the image more if you want to see them.
Now what about doubling the focal length of the lens ? You add an octave of detail on the subject (Christmas ornaments), but do you add an octave of noise ?
Don Cox
I've never saw term for DR of single pixel.
DR is term used for entire sensor but for practical reasons it's always measured over some small subarray of pixels as average DR of multiple pixels. There is reasonable expectation that pixels behave similarly across entire sensor.
DR is given as FWC (Full Well Capacity)/RN (Read Noise), but FWC is here for average pixel in the measured subarray and not some specific pixel and the same is true for RN.
Difference of pixel size isn't so large. It's 11 microns^2 (Oly) vs 14 microns^2 (Sony).
Also at ISO400 Oly is saturated at 16K e- and Sony at 8.6K e-. Because PDR works with shot noise which is dependent on saturation, I'm not so surprised that that difference isn't so big.
Architeuthis
Leading Member
- Messages
- 653
- Reaction score
- 333
Hi DiMachi,
I wonder how you arrive at the pixel areas you cite. Calculated from pixel pitch the pixels of A7RIII have approx. 2x the area (and FWC) and A7III approx. 3.5x:
Not all of this area will be used for light gathering, since there is also some "dead" surrounding of a pixel, the ratio circumference/area will, however decrease, the bigger the pixel becomes. BSI also will increase the area used for light gathering...
edited after posting: Sorry DiMachi, I just see that in the original link ther is EM1II with A7RIV compared - I had originally RIV, RIII and III in the graph, decided to leave only RIII, but erroneously let RIV...
My mistake - I apologize!
Still there is the (apparent) dicrepancy that larger pixel area does not help to improve DR...
Wolfgang
Last edited:
Iliah Borg
Forum Pro
We don't have access to FWC in a vast majority of cameras because close to FWC characteristic curve is non-linear, and fixed pattern noise becomes problematically high (4 .. 5% of FWC, easy, and that is very visible) , so the manufacturer clips the signal, making ADC saturation level lower than sensor saturation level by about 1/2 to 1 stop.
What you can access is limited by design, and it is the max level in raw data (clipping in raw).
Using this number as the ceiling is, strictly speaking, also wrong without additional verification. This value may be too high, to accommodate for some of the photon shot noise. That's why Canon are explicitly, in metadata, specifying the specular white level and linearity upper margin, both being lower than the value for clipping in raw.
It doesn't, FWC is about a single pixel.
Funny, I wanted to apologize to you Wolfgang as I've used data for A7RIV and not A7RIII and I already had message prepared for post but DPR logout me before I did that.Hi DiMachi,
I wonder how you arrive at the pixel areas you cite. Calculated from pixel pitch the pixels of A7RIII have approx. 2x the area (and FWC) and A7III approx. 3.5x:
Not all of this area will be used for light gathering, since there is also some "dead" surrounding of a pixel, the ratio circumference/area will, however decrease, the bigger the pixel becomes. BSI also will increase the area used for light gathering...
edited after posting: Sorry DiMachi, I just see that in the original link ther is EM1II with A7RIV compared - I had originally RIV, RIII and III in the graph, decided to leave only RIII, but erroneously let RIV...
My mistake - I apologize!
Still there is the (apparent) dicrepancy that larger pixel area does not help to improve DR...
Wolfgang
Now I see your apology!
So apology is mutual!
Now to the discussion.
At ISO400 saturation for Sony A7RIII is 13K e- and not 8.6K e- (that's value for A7RIV). So then difference between Oly and Sony isn't so big but considering all the facts, I still think that PhotonstoPhotos results seems reasonable.
One reason is that ISO100 (or any specific ISO) does means slightly diffent sensitivity for one manufacturer than for other one. That means that graphs for different cameras can have some offset in the horizontal axis.
BSI also doesn't neccessary mean that sensor has better parameters. Well designed and manufactured FSI sensor with microlenses can be better than not so well designed and manufactured BSI sensor.
Last edited:
They do not, in general. A larger pixel has to count more photons, and as such it usually has a higher input referred read noise (in e-). See here (ISO 100). As an analogy, imagine a preamp and a power amp. In absolute units, the power amp would add more noise but that would be OK because the output would be much stronger. With typical signals, the SNR could be the same.
bclaff
Forum Pro
If you used a much darker patch to start then the cropped and expanded version would show more apparent noise.
bclaff
Forum Pro
Yes. (And DOF will appear more shallow too.)Or the closer you get to a noisy photo the noisier it will look ?
bclaff
Forum Pro
Glad you sorted that out.
I call single pixel dynamic range Engineering Dynamic Range (EDR) since this is what you will often find on the actual specifications sheet for a sensor.
Not strictly sensor size but that's an OK first approximation.
Technically what matters is the shape of of something called the Photon Transfer Curve (PTC). Here's the A7R3 for example:
The quick area approximation essentially assumes a straight line with a slope of 1/2 like in the middle of the curve.
Because pixel size is largely irrelevant; what matters is the total area collected (for something normalized like PDR).
--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )
A better analogy is raindrops falling into buckets to measure rainfall. If you put out a 10cm x 10cm bucket, and you have roughly 200 raindrops per 100 square centimeters over the unit time you're measuring, you might get 109 raindrops or 84 raindrops, based on luck.
If you put out four 5cm x 5cm buckets, you'll still capture the same total number of raindrops, and your noise will be roughly the same once you add them up. There might be a little more with the smaller bucket -- it's easier to measure 100 raindrops than 25, so whatever measurement errors your tools introduce might change things, but at the end of the day, it's more-or-less the same thing if you can measure well.
We can measure increasingly well.
If you put out four 5cm x 5cm buckets, you'll still capture the same total number of raindrops, and your noise will be roughly the same once you add them up. There might be a little more with the smaller bucket -- it's easier to measure 100 raindrops than 25, so whatever measurement errors your tools introduce might change things, but at the end of the day, it's more-or-less the same thing if you can measure well.
We can measure increasingly well.
With sensors, measuring accuracy typically increases (in raindrops) for small buckets. In the end, the result is more or less the same.
Great Bustard
Forum Pro
- Messages
- 45,961
- Solutions
- 17
- Reaction score
- 34,046
Same difference.
OK -- I'll rewrite it:
Consider two photos of a scene from the same position. One at 100mm f/5.6 1/400 and the other at 50mm f/5.6 1/400 and cropped to the same framing as the 100mm photo. Any given proportion of the 100 photo will be made with 4x as much light as the crop from the 50mm photo, and thus be less noisy with more DR. It will also have higher resolution since it is made with 4x the number of pixels.
Similar threads
Keyboard shortcuts
- f
- Forum