Relating machine vision camera sensitivity to ISO ratings

[JMC]

I'm trying to get my mind around something and need some advice on it please.

 

The question is this. How can I compare the sensitivity of high street cameras with ISO ratings, to cameras such as the ARTCAM-2020UV-USB3 camera with the GSENSE2020BSI sensor as discussed here - https://www.ultravioletphotography.com/content/index.php/topic/4143-a-first-test-with-my-gsense2020bsi-camera/page__view__findpost__p__39255

 

What I'm trying to figure out is whether moving to something like the Artcam will be significantly better for my work at and below 300nm than my monochrome Nikon cameras, especially for the microscopy?

 

So for instance as an example, if I need to turn up the ISO on my Nikon d800 monochrome to 12,800 to get an image with a 10s exposure at 310nm, would the Artcam image be more or less noisy than the Nikon one to capture the same image? I get the resolution will be very different, it's more about the noise.

 

Common sense tells me that that sensors themselves wont be hugely different in sensitivity (apart from the BSI/FSI difference), so it'll then come down to how the information from the sensor is processed.

 

Hopefully I'm explaining this clearly enough, and any advice is welcome.

[Andy Perrin]

Jonathan, I would love to know the answer to this. My situation with the TriWave is even worse: the camera has an unknown sensitivity curve that in some cases even departs from what the company published, and on top of that it’s analog, so I have

1) analog gain (maybe comparable to ISO) of “1/2, 1, 2, 4” that I can set with the TriWave software

2) the gain of my frame grabber

 

So what is my equivalent ISO? WHO KNOWS?

[Stefano]

Do you use the same type of grabber that is used to digitize VHS tapes? Just a curiosity.

[nfoto]

On a general note, smaller sensors tend to be more noisier than the bigger sizes.

[Stefano]

Also, it may be obvious but often pixel size (area) matters, but there are also other factors involved.

[Andy Perrin]

Do you use the same type of grabber that is used to digitize VHS tapes? Just a curiosity.

In the general sense (since it goes NTSC -> digital) but I use a very high-end one.

 

The digitizer is 8-bit, but by averaging large numbers of frames, it's possible to beat this limitation and get a higher effective bit depth, as described here:

https://www.cypress.com/file/236481/download

[dabateman]

Jonathan it doesn't work out directly that way.

Things you need to know.

Art measure, the software that runs this camera has global gain setting, which can be set from 0 to 255. The default value is 64. Maybe best at that setting.

 

The 2020bsi has a saturation capacity of 54ke- and read noise is 1.6e- at 12 bits.

 

Now you can read through these threads

 

https://www.cloudynights.com/topic/565556-how-to-translate-dslr-iso-into-ccd-gain/

 

https://www.cloudynights.com/topic/565620-question-about-gain-setting-on-cmos-cameras/#entry7688474

 

But basically you can run a faster shutter speed than your Nikon.

 

For my piHQ the base analog gain seems to be roughly equal to 80 ISO. I typically like -ag 4, roughly 320 ISO, but since this is analog gain not digital gain, there isn't noise.

 

What I love about the pi camera is that you have analog gain and digital gain separate. You can float the two to balance out an exposure setting or just play with one. Analog gain can be set from 1 to 16. Digital gain adds noise.

[Andy Perrin]

That sounds exactly backwards David. Digital gain is just multiplication by some number so can’t add noise. Analog gain adds noise because amplifiers have thermal fluctuations.

[dabateman]

Analog gain is before the analog to digital converter in the camera. Like changing your base ISO of the camera. Newer cameras have dual ISO base values to take advantage of this lower noise trick. The problem is less dynamic range. So not good to ramp up analog gain to the max.

 

Digital gain is like ISO in the non dual ISO cameras. Its after the ADC and just adds noise as you digitally boost the signal and noise from ADC.

 

I may not be explaining it clearly, as I am using old and new and ISO in the mix.

 

Think if you expose with short shutter speed you don't have much light. So if you digitally boost that to increase brightness now you increased digitally the harder to see background noise. So digital will be noisy. So digital amplifiers all losses in the system.

 

[Andy Perrin]

Digital gain doesn't add any noise. The noise that is already present gets bigger, but that's not adding any frequencies that were not already present. Multiplication by a constant can't add new frequencies. Both the image and the noise are increased by the same factor, so the signal-to-noise ratio is the same.

 

Analog gain (the way ISO used to be handled on older cameras before they started making it a hybrid) adds other frequencies to the mix from thermal and other effects prior to the ADC even touching the signal. So the noise already present is amplified plus new noise added from the amplifier.

[dabateman]

Andy this explanation is better than mine:

https://www.edmundoptics.com/knowledge-center/application-notes/imaging/basics-of-digital-camera-settings-for-improved-imaging-results/

 

"Gain is a digital camera setting that controls the amplification of the signal from the camera sensor. It should be noted that this amplifies the whole signal, including any associated background noise. Most cameras have automatic gain, or autogain, which is abbreviated as AGC. Some allow the user to turn it off or set it manually.

 

Gain can be before or after the analog-to-digital converter (ADC). However, it is important to note that gain after the ADC is not true gain, but rather digital gain. Digital gain uses a look up table to map the digital values to other values, losing some information in the process.

 

Gain before the ADC can be useful for taking full advantage of the bit-depth of the camera in low light conditions, although it is almost always the case that careful lighting is more desirable. Gain can also be used to ensure that the taps of multi-tap sensors are well matched. For a detailed discussion of sensor taps, view Imaging Electronics 101: Camera Resolution for Improved Imaging System Performance. In general, gain should be used only after optimizing the exposure setting, and then only after exposure time is set to its maximum for a given frame rate. "

[Andy Perrin]

David, nothing in that contradicts anything I said, or changes my understanding of what’s going on? Except this statement which maybe needs to be qualified by the author:

“Digital gain uses a look up table to map the digital values to other values, losing some information in the process.”

 

There shouldn’t be any loss of information in applying digital gain unless you store your image at 8 bits/channel. When I work in MATLAB, I store my images at 64 bits so that any round off error is negligible.

——

 

I’m honestly losing track of this discussion at this point though. What on earth were we originally talking about? I think Jonathan wanted to know how to compare ISO settings in regular cameras to gain settings in scientific cameras. Why not use the PSNR to match measurements of both cameras and try to match the settings up? Then make a little table, “ISO100 = gain xxx” etc.

[dabateman]

Back to Jonathan point. I think he has a Nikon that can see UVC. But it needs an exposure of say F8, 10 seconds at ISO 6400. I think he wants to understand how many more stops is he gaining going to an Artray UVC camera where its not clear from the gain unit of 64 being used. So to guess how fast or what NA objectives can he get away with.

The problem is you can't really know until you test it yourself. We don't know the read noise and saturation points at the specific wavelengths so will be hard to calculate.

 

So my easy answer is that it will be faster and will be able to use a higher NA objective. But is it worth the cost? That will be up to you. So will pay $3000 for 5% improvement in lens sharpness. So this is similar. I like the pi, even though there is a steeper learning curve.

 

 

Andy, back to our side conversion. The problem is digital gain is not just a multiplier. You have losses. Its comparable to color white balance correction from raw file with all the bits (analog gain) versus from a jpeg image where you threw out a bunch of information and added noise from the ADC (digital gain). Thats best I can think of now as a comparison.

[JMC]

Ok, so it's complex then. At least I wasn't getting confused by something simple.

 

Yes, I can see UVC with my Nikon d850, but the Artcam comes with a sensor which has been optimised for UV. The quantum efficiency at 300nm and below looks very different to a normal sensor.

 

The Nikon I can increase the ISO and still get usable results. But is there a significant improvement in real term sensitivity by moving to the Artcam - enough of an improvement to justify buying one? Obviously I'll lose resolution, but that is not a huge issue for me, especially if sensitivity is dramatically improved.

[Stefano]

Maybe we should look at quantum efficiency too?

[dabateman]

I would hope that the Artray using the GSENSE 2020BSI sensor which has been recently used to detect soft x-rays

 

https://www.gpixel.com/news/gpixel-announces-new-pulsar-technology-enabling-gsense-products-for-vuv-euv-soft-x-ray-and-electron-direct-detection-imaging/

 

Is at least as sensitive as the MaxMax monochrome converted pi HQ camera with 12Mpixels. The pi HQ can record video in UVC using my filter (same one Jonathan has) with quartz lens (Jonathan also has) so thats at 1/30 frames per second with little noise, that I could frame grab shots of my finger.

Compared to my recently discovered Olympus Em5mk2, which can see UVC at IS0 800/1600 with F4/5.6 and 8 seconds.

 

So I would say for sure Jonathan the Artray will be faster. They didn't have the software offer any shutter speed slower than 1 seconds, as wasn't needed. But Wil has a beta version extended to 8 seconds of the software as they seem to work with their customers.