-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Interesting read on light limits in digital photography
- Thread starter PJPfeiffer
- Start date
HI, thanks for posting this. It promises to be an interesting and thoughtful read for an ignoramus like me
ZodiacPhoto
Senior Member
I already wrote about this approach in another thread, but this is the only way to higher DR in a single exposure than current sensors.
It relies on ability to place more electronic circuits on the back of the sensor, thus increasing the speed at which the pixels can be read – see this Sony technology:
https://www.dpreview.com/news/56961...phone-sensor-that-can-shoot-1000-fps-at-1080p
In high-contrast cases, some pixels will be fully discharged (clipped highlights), and some – not discharged enough to measure (clipped shadows).
Time-domain acquisition is a paradigm – shift: measure not just the remaining photocell voltage, but also a TIME that it took for the cell to fully discharge, if it was less then exposure time. From there, you can easily calculate the highlight value that would be otherwise clipped.
If your shutter time was 1/100, and some pixels clipped at 1/500, some – at 1/423, 1/327, 1/248, etc., that will give you the would be value for that pixel. You can also wait longer for shadow pixels to register a minimal discharge, say, 1/64, 1/50, etc., to calculate shadows value.
It is easier and more precise to measure in time domain than in analog domain.
Another interesting approach is dual-resolution sensor.
Imagine a FF 36MP sensor, where the central area of a size of Micro 4/3 sensor is made of 4 pixels for each of FF pixels. In FF mode, these smaller pixels are binned together, and the whole sensor works like a 36MP FF sensor.
Then you can switch to M4/3 mode, use the central part of your lens image, where the lens is the sharpest, and get an equivalent of a 2X tele-converter!
This is to illustrate that sensor technology as a whole is not nearly at the dead end.
It relies on ability to place more electronic circuits on the back of the sensor, thus increasing the speed at which the pixels can be read – see this Sony technology:
https://www.dpreview.com/news/56961...phone-sensor-that-can-shoot-1000-fps-at-1080p
In high-contrast cases, some pixels will be fully discharged (clipped highlights), and some – not discharged enough to measure (clipped shadows).
Time-domain acquisition is a paradigm – shift: measure not just the remaining photocell voltage, but also a TIME that it took for the cell to fully discharge, if it was less then exposure time. From there, you can easily calculate the highlight value that would be otherwise clipped.
If your shutter time was 1/100, and some pixels clipped at 1/500, some – at 1/423, 1/327, 1/248, etc., that will give you the would be value for that pixel. You can also wait longer for shadow pixels to register a minimal discharge, say, 1/64, 1/50, etc., to calculate shadows value.
It is easier and more precise to measure in time domain than in analog domain.
Another interesting approach is dual-resolution sensor.
Imagine a FF 36MP sensor, where the central area of a size of Micro 4/3 sensor is made of 4 pixels for each of FF pixels. In FF mode, these smaller pixels are binned together, and the whole sensor works like a 36MP FF sensor.
Then you can switch to M4/3 mode, use the central part of your lens image, where the lens is the sharpest, and get an equivalent of a 2X tele-converter!
This is to illustrate that sensor technology as a whole is not nearly at the dead end.
PJPfeiffer
Senior Member
The article really doesn't state sensor tech is dead ... but rather is discussing the light limits and noise.
I read just a few paragraphs, too many mistakes.
1. The binning thing, as already pointed out, is wrong
2. "In this case, especially, the Nyquist-Shannon theorem is not fully applicable. The theorem refers to point sampling of a continuous image. But pixels are not points: they sense average light intensity over a finite area. Especially when the high frequency signal one is trying to render is an “overtone” of a lower frequency fundamental (such as an attempt to make a sinusoidal signal into a square wave), this averaging over a finite space makes it easier to detect “beyond-Nyquist” frequencies."
Exactly the opposite. The sampling theorem is still applicable but to the image convolved with a pixel. This does not change the Nyquist limit but attenuates the high frequencies close to it making them more sensible to noise. Of course, the AA filter is designed to work with pixels, and the combined effect is intended.
3. So the practical “speed limit” for 35mm cameras is not that much higher than the current level of about 24 megapixels for a perfect f/10 lens.
Demonstrably wrong.
4. White light. In this case, we would have three concentric Airy disks: one smaller in diameter for blue light, and one larger for red light.
Pure nonsense. He thinks that the spectrum of white light consist of three monochrome signals. We all know that we do not see three Airy disks (and they are infinitely many anyway).
5. This filtering reduces the modulation transfer function of a sensor at spatial frequencies close to or slightly below the limit of that the number of pixels can render.
A sensor does not have MTF; it is a sampling device transforming a continuous image into data on a finite grid.
6. All pure wavelengths are a mix of at least two primary colors
Huh? What are pure wavelengths - spectral colors? In what ways are they a mix - spectral mix or they appear as the same colors to us?
7. His analysis of Bayer filters and AA filters is naïve at best.
8. Technology today is approaching or even at the limits of resolution, film speed, and image size imposed by the laws of physics for cameras in the 35mm format or smaller.
LOL! The article was written in 2009. My 5D2 must then be state of the art today.
9. Bigger equipment size/sensor size means better pictures ... but ... Therefore there is no real advantage to the large format.
1. The binning thing, as already pointed out, is wrong
2. "In this case, especially, the Nyquist-Shannon theorem is not fully applicable. The theorem refers to point sampling of a continuous image. But pixels are not points: they sense average light intensity over a finite area. Especially when the high frequency signal one is trying to render is an “overtone” of a lower frequency fundamental (such as an attempt to make a sinusoidal signal into a square wave), this averaging over a finite space makes it easier to detect “beyond-Nyquist” frequencies."
Exactly the opposite. The sampling theorem is still applicable but to the image convolved with a pixel. This does not change the Nyquist limit but attenuates the high frequencies close to it making them more sensible to noise. Of course, the AA filter is designed to work with pixels, and the combined effect is intended.
3. So the practical “speed limit” for 35mm cameras is not that much higher than the current level of about 24 megapixels for a perfect f/10 lens.
Demonstrably wrong.
4. White light. In this case, we would have three concentric Airy disks: one smaller in diameter for blue light, and one larger for red light.
Pure nonsense. He thinks that the spectrum of white light consist of three monochrome signals. We all know that we do not see three Airy disks (and they are infinitely many anyway).
5. This filtering reduces the modulation transfer function of a sensor at spatial frequencies close to or slightly below the limit of that the number of pixels can render.
A sensor does not have MTF; it is a sampling device transforming a continuous image into data on a finite grid.
6. All pure wavelengths are a mix of at least two primary colors
Huh? What are pure wavelengths - spectral colors? In what ways are they a mix - spectral mix or they appear as the same colors to us?
7. His analysis of Bayer filters and AA filters is naïve at best.
8. Technology today is approaching or even at the limits of resolution, film speed, and image size imposed by the laws of physics for cameras in the 35mm format or smaller.
LOL! The article was written in 2009. My 5D2 must then be state of the art today.
9. Bigger equipment size/sensor size means better pictures ... but ... Therefore there is no real advantage to the large format.
PJPfeiffer
Senior Member
Thank you for clarifying for the membership here
Perhaps contacting the author is also a good idea
Perhaps contacting the author is also a good idea
Keyboard shortcuts
- f
- Forum