Digital photography ISO measurement, why uniform across whole image?

Question

ISO is been around since analogue photo times. Any, it should have been uniform for obvious reasons - you could never know what will be captured in each part of the film. So, you go uniform (ISO 100, ISO 200), select appropriate one for given conditions and do your best. Fast forward to today and we have same approach. Despite the fact that in digital photography "you know" or, rather, image sensor knows what will be captured by each pixel or group of pixels. Here's the question, why it couldn't measure sensitivity per area or, even, per pixel and have nonuniform ISO settings across the sensor? What prohibits having ISO level automatically set per pixel?

Answer 1

Stop and think about what you are saying. Assuming one were willing to spend the prohibitive cost to do what you are proposing, what would the result be? If every pixel were adjusted using ISO so that it is exposed "properly" you would wind up with an image where every pixel is the same brightness! There would be zero contrast. None. You would not be able to see details of anything that is a uniform color. And don't even think about going monochrome! You would wind up with an image that looks like an 18% gray card.

hm... maybe i've formulated incorrectly if that's the perception. I'd rather imagine the results along the HDR-like outcome where highlighted areas would have 100 ISO applied and dark areas would be treated with, maybe, ISO 800 while mid one would get made with ISO 200. in other words, as HDR is achieved by combining outcomes through altering exposures - in ISO based scenario it would be achieved through combining multi-ISO application.

The purpose of HDR isn't ultimately to reduce noise. It is to take a scene with a wider dynamic range than can be captured by current cameras and/or displayed by current display mediums and squeeze that additional dynamic range into the space allowed by the display medium. Increasing ISO doesn't increase dynamic range. It decreases dynamic range because it takes half as many photons to reach the equivalent of full well capacity.

HDR, at least in a digital environment to which you seem to be referring, is much more complicated and involved than what you state. Ultimately the tone mapping that must be done is more a manipulation of local contrast versus overall contrast. There would be little to gain by shooting the bright areas at ISO 100 and the dark areas at ISO 800 - because the areas in which ISO 800 creates the most noise are the shadows! Brighter areas look perfectly fine at ISO 800. If you amplify one part of the picture more than another, then you still have the problem that some parts of the dark areas that are still darker than the darkest parts of the bright areas will look brighter than those parts of the bright areas.

agreed, it is much more complicated process that i could ever dare to represent. However, it is still exposure based for sourcing the information to work with. I'm simply saying that while i generally understand aperture and exposure as relevant and applicable in unified way to overall image capture i can't (yet) to understand reasons for ISO being applied the same. In film - it is clear for obvious reasons cos otherwise you'd need to create dedicate film for each capture. But for digital - there's still something that goes against the logic in my mind.

To take advantage of various ISO sensitivities on discrete parts of the sensor you'd have to find a way to alter the exposure time on a pixel by pixel basis. So now you are requiring sensors that can provide global electronic shutter. Even when the entire sensor collects photons for a universal time period they are very expensive to produce. The computing power to do what you propose would raise the cost of such a camera astronomically. Creating a sensor chip that could follow the instructions resulting from such computations would probably exceed the budget of most third world countries.

Answer 2

ISO of a sensor involves what voltage it is running at. To modify this in just an area or pixel would require a lot of complex wiring (how can the manufacturer guess that you want to increase the ISO of pixel 432,288) and a lot of added interfacing and programmin. In simple terms, such a capability would be a very costly feature rarely used.

Answer 3

I think you may be asking about uniformity of response, in which case most if not all cameras have built-in correction tables to remove dark offsets for each pixel and gain differences across pixels. This guarantees that the signal output from every pixel for N input photons is the same.

If you're asking about adjusting ISO so dark parts of the scene get more gain, then you'd be asking for dynamic gain adjustment on a per-pixel basis. Even then, as Michael Clark pointed out, you'd end up with at best a very washed-out image.

If you have a scene with huge contrast (dark to brightest), there is another option. Expensive, but available: sensors with logarithmic amps. This gives the pixels a huge effective dynamic range, and you can then post-process the log scale output to do whatever you want.