Sensors, pixels and histograms

[David Sutton]

Hi David,

given the many responses to have had here, how have your view on the topic at hand change if at all. Are you able to define the elements on a sensor that captures the light and their relation to how they are presented in an image program/ raw converter?

When you look at a histogram - pixel, photo-sensor - all its telling you is the relation of the pixels in terms of density within 0-255 (8bit) or the equivalent in 16bit (RGB histogram gives you information of how each color is distributed and if clipping has occurred which colors are clipped - which in turn can help you address the issue or live with it. So, if you look at a histogram and seeing something else, then please tell, cos I would love to know.

Please enlighten me, I am just as curious as you

thanks

Henrik

Hi Henrik. I'm currently doing one last photo trip before the end of my summer holidays. At the end of each day I've been plugging the vodem into my laptop and reading and replying to these posts. Some technology is both cheap and really cool.
While on the  road (or water) I've been discussing the responses with my travelling companion and I will be very interested to see the resulting images on a large screen when I get home tomorrow, as my ideas about image making shifted today. I'll post then.
Best wishes, David

[David Sutton]

Hi David,

given the many responses to have had here, how have your view on the topic at hand change if at all. Are you able to define the elements on a sensor that captures the light and their relation to how they are presented in an image program/ raw converter?

When you look at a histogram - pixel, photo-sensor - all its telling you is the relation of the pixels in terms of density within 0-255 (8bit) or the equivalent in 16bit (RGB histogram gives you information of how each color is distributed and if clipping has occurred which colors are clipped - which in turn can help you address the issue or live with it. So, if you look at a histogram and seeing something else, then please tell, cos I would love to know.

Please enlighten me, I am just as curious as you

thanks

Henrik

Hi Henrik. Wow, your questions put me on the spot a little, which is probably a good thing, so here is my attempt at an answer. What I have learned is not new for a lot of people, but photography is such a personal thing that what I do with this information may result in something quite different from someone else's approach. The images we make say a lot about the us and often reveal as much of the photographer as the photographed. I use a camera for my own pleasure and as as a means of exercising my imagination and skills, and after printing they are usually shown once and then put in a drawer.
Okay, first the larger picture. I'm even more convinced the words we use to describe what we are doing are powerful tools that limit or expand our horizons. Thinking about this while driving to the next shoot, and how I choose to think about image capture, I also realised that what I choose to remember in my life will define what sort of life I have. Good and bad things happen and we learn from both, but if I have a life filled with happy memories it will be because I have chosen my attitude to external events and chosen what to recall.
Good, that's out of the way. Next, this discussion has defined the difference for me between photos and images. A photo for me is what happens when you are walking along with a camera and think “That looks interesting”. Click. When I look at the histogram I am treating it as a measure of exposure and leave it at that. Images are what happens when I have a mental picture and I want to turn it into a print (“what I saw” versus “what I see” I guess). I go looking for source material (input) to turn into pixels. When I look at the histogram I try to see what parts of it correspond to what is in front of me. In the bits of interest what is the signal to noise ratio? What has clipped in the shadows and highlights and does it matter? Knowing the histogram is of a jpeg generated from the raw file, I want to know how close that is to “reality”. On my camera, if I set the colour space to Adobe rgb I get a fair idea of where the highlights will clip and the information in that channel will be lost. If I set it to srgb it's better for showing shadow clipping. And there is uniwb as a custom setting if I need it.
I've tried to find some examples and how I think about them now. Here is a photo of a duck:
[attachment=19711:186NoTextVFAPRel.jpg]
I wanted to show the determination in this little fellow, so I cropped wide and sharpened the water to show him pushing against something strong, and sharpened his eyes to show he wasn't fazed by that. Printed on Epson Velvet Fine Art paper to give the water more substance and texture. Not much else really.

Here is an image of a lighthouse:
[attachment=19712:_MG_6838...nd_uprez.jpg]
Lighthouses really interest me. One of the first photos I took over 50 years ago when I was about 6 years old was a lighthouse. I still have the negative. I am slowly doing a series and want to show in the final prints something of the reason they have ended up as such strong symbols. I wanted to show one in relation to the size of the surrounding ocean, but I haven't got any vast ocean bits so here cropped to show a big sky instead. It was taken in the Western Isles off Scotland, which in my imagination is a place of mystery. I wanted to look at it and ask if I really went there or dreamed it, so to get that look I stripped out a lot of the information from the image by heavy cropping and printing large on Velvet Fine Art paper so the sky looked like it had been brushed on with watercolour. There was a little sharpening on the lighthouse to give it more reality. Looking at the histogram, I wanted to shift it to the right to maximise my signal to noise ratio, so when I was left with just a little amount of information in the sky, it wouldn't be mainly splodgy noise. I didn't care how much of the histogram showed clipping, as long as it wasn't in the bit of the image I wanted to  use. Some guesswork here so I did take a few shots to be sure.

Finally here are some images of sheep with  trees taken from a series motivated by my dislike of how the female form is Photoshopped in fashion photographs:
[attachment=19713:_MG_2698Border.jpg][attachment=19714:_MG_3345...elBorder.jpg]

 Looking at the histograms I wasn't too worried about shadow noise as I was going to send a lot of the print to black and would probably end up adding noise anyway. I was most concerned by the bits of sky and where the right side of the histogram was sitting. I hate having bits of sky in an image as they can go to white in a print and it shows, but I wanted to have information in the wool.
I find symmetry in an image disturbing and often repellent and I don't like it, though I guess most people don't agree. So if I want to unsettle myself I put some symmetrical bits in a print. I think sheep and these trees are really spooky, so I made the whole thing almost symmetrical and painted in light and shadow.  In the colour print I painted in some desaturation as well. They were printed on Harman gloss, which I use when I want to make a print look “hyper real”.

What I want to know now is how much does shifting the histogram to the right during image capture affect the signal to noise ratio on my camera, and how much extra information have I recorded on my camera? I intend to shoot a scene with the histogram  just short of clipping, then repeat with it touching the three quarter mark and then touching the half way mark, equalise the exposure and then do some aggressive editing in Photoshop to see which version falls apart first and see if it is visible in print. I want to know not only what works with equipment, but also at what point is it likely to fail. I think I'll need to take the above advice and learn DCRaw and rawanalyse, so this is a project for later in the year.
David

[Daniel Browning]

What I want to know now is how much does shifting the histogram to the right during image capture affect the signal to noise ratio on my camera, and how much extra information have I recorded on my camera?

It depends on the tonal level you are looking at. For example, say we're looking at a tonal level that is 3 stops below saturation in the first raw file at ISO 100. If you increase exposure by 2 stops, that tonal level will now be 1 stop below saturation, and the SNR will have doubled. (A one stop increase in exposure increases SNR by 41.1%.) If you leave exposure alone and increase ISO by 2 stops (ISO 400), the tonal level will now be 1 stop below saturation, but there is no change whatsoever in SNR. Both of these facts are because the only noise in that tonal level is photon shot noise.

As another example, say we're looking at a tonal level that is 8 stops below saturation at ISO 1600. Here the total noise power will be dominated by read noise instead of photon shot noise. If you increase exposure by 2 stops, the tonal level will now be 6 stops below saturation, and the SNR will have improved a factor of four (almost). If you increased exposure by two stops and decreased ISO by the same amount, then the tonal level and histogram would remain the same, but the SNR would still improve; but this time it would be a bit less improvement with certain cameras, because some CMOS sensors with analog gain have much less read noise (relative to any given tonal level) at higher ISO.

Hope that helps.

[bjanes]

What I want to know now is how much does shifting the histogram to the right during image capture affect the signal to noise ratio on my camera, and how much extra information have I recorded on my camera? I intend to shoot a scene with the histogram  just short of clipping, then repeat with it touching the three quarter mark and then touching the half way mark, equalise the exposure and then do some aggressive editing in Photoshop to see which version falls apart first and see if it is visible in print. I want to know not only what works with equipment, but also at what point is it likely to fail. I think I'll need to take the above advice and learn DCRaw and rawanalyse, so this is a project for later in the year.

As far as shot noise goes (and shot noise predominates in all but the extreme shadows, where read noise predominates), it is the number of photons captured and not the appearance of the histogram that counts. If the histogram is to the left and you move it to the right by increasing ISO, you have not collected any more photons and the shot noise will be the same. However, if you can't increase actual exposure because of shutter speed or depth of field considerations, then upping the ISO will decrease read noise up to a certain ISO, at which point there are diminishing returns. Beyond that point, you merely decrease highlight headroom by increasing ISO as Emil Martinec explains in detail. That point of diminishing returns varies with the camera. The the Nikon D3 it is at about ISO 800, but with the D3x, it occurs at about twice base ISO.

If you don't mind having a dark preview on the LCD in this case, it is better to increase "exposure" in the raw converter and avoid blowing the highlights.

[David Sutton]

It depends on the tonal level you are looking at. For example, say we're looking at a tonal level that is 3 stops below saturation in the first raw file at ISO 100. If you increase exposure by 2 stops, that tonal level will now be 1 stop below saturation, and the SNR will have doubled. (A one stop increase in exposure increases SNR by 41.1%.) If you leave exposure alone and increase ISO by 2 stops (ISO 400), the tonal level will now be 1 stop below saturation, but there is no change whatsoever in SNR. Both of these facts are because the only noise in that tonal level is photon shot noise.

As another example, say we're looking at a tonal level that is 8 stops below saturation at ISO 1600. Here the total noise power will be dominated by read noise instead of photon shot noise. If you increase exposure by 2 stops, the tonal level will now be 6 stops below saturation, and the SNR will have improved a factor of four (almost). If you increased exposure by two stops and decreased ISO by the same amount, then the tonal level and histogram would remain the same, but the SNR would still improve; but this time it would be a bit less improvement with certain cameras, because some CMOS sensors with analog gain have much less read noise (relative to any given tonal level) at higher ISO.

Hope that helps.

Yes it does help. I understand the signal to noise ratio is proportional to the signal power (S/N α P if I remember my maths). Most noise I'm used to seeing is read noise. I think I've just learned that shot noise is caused by the quantisation of light, and thus random variations in the number of photos arriving at the receptor at any given moment. And the shot noise is also inversely proportion to the signal frequency (S/N α 1/f). As the frequency goes up, the photon energy increases and the number of photons arriving at any moment for that frequency decreases. So the shot noise will increase as the histogram is moved to the right. I think I've got that right. I don't recall seeing it in print. Just had a look at a blue sky from a raw file at about 3:1 I can clearly see noise in there around the ¾ tones. That must be it.

Edit: bjanes you post got in ahead of me. Thanks for the clarification