I know you've had several suggestions on hot pixels, but I've had an idea bouncing around in my head for a long while and I think it might actually turn out to be highly useful. My thought is that generally speaking a large amount of the "noise" in an image is a result of the varying sensitivities in the CCD as well as the interpolation scheme used to extract the data from a Bayer pattern (not related) sensor.
My hypothesis is that if you built a profile out of several light (not fully white) images of different exposure length (maybe with a fixed white balance) and several several dark images also with different exposure lenghts) that you could build a model (for a given camera) of how the sensor pixels vary. The user might have to input the exposure setting of each image (if EXIF wasn't present.) In the case of really long exposures (where the pixels max out), you'd have to do the standard sort of stuck pixel removal games.
The difference comes in situations where the pixels haven't maxed out. In these cases you could provide a slider that moves the correction in exposure time to reduce the brightness of the underlying pixels (possibly only in single color channels). To handle jpeg images you'd also have to deal with some of the surrounding pixels, but I think that should be doable.
You might also have to provide a bias control for dealing with color/white balance effects. But I think with some refinements this process could yield results far in advance of your (admittedly amazingly good) detection/removal schemes for noise. That said this feature would only be used by a subset of your customers who work primarily with digital photography. However, if it was done right, I think you might gain lots of new customers.
Hot pixel correction suggestion (not exactly removal)
Thank you for the suggestion.
Yes, this kind of scheme is very much possible and moreover, it is already used in certain fields related to image processing. It is not so much popular in digital photography probably because the effect of varying sensitivies in the CCD is weaker than the ordinary electronic noise of the circuitry, especially when amplified in high sensitivity modes. If the electronic noise was eliminated then the unevenness of the CCD sensitivity would probably be more noticeable. As is it right now, we only see unevenness of the CCD in the form of hot and dead pixels but this is not exactly unevenness of sensitivity.
Yes, hotpixels remain a problem in digital photography and newest digital cameras try to solve this problem by using an internal map of hot pixels. On the other hand, I am not aware of any digital camera that would take into account the unevenness of sensitivity, at least the type of unevenness we are discussing here.
So, I agree with you on the point of feasibility but I am not that sure that the effect would be noticeable. Still I am aware of applications of CCD imaging where the effect would be important.
Vlad
Yes, this kind of scheme is very much possible and moreover, it is already used in certain fields related to image processing. It is not so much popular in digital photography probably because the effect of varying sensitivies in the CCD is weaker than the ordinary electronic noise of the circuitry, especially when amplified in high sensitivity modes. If the electronic noise was eliminated then the unevenness of the CCD sensitivity would probably be more noticeable. As is it right now, we only see unevenness of the CCD in the form of hot and dead pixels but this is not exactly unevenness of sensitivity.
Yes, hotpixels remain a problem in digital photography and newest digital cameras try to solve this problem by using an internal map of hot pixels. On the other hand, I am not aware of any digital camera that would take into account the unevenness of sensitivity, at least the type of unevenness we are discussing here.
So, I agree with you on the point of feasibility but I am not that sure that the effect would be noticeable. Still I am aware of applications of CCD imaging where the effect would be important.
Vlad
I'm not sure if that's really true, that the effect would be unnoticable. I should do a bit more research in this I suppose. My sneaking suspicion is that a lot of the chromatic noise is a result of variations in the sensitivity and the thickness of the filter (RGBG) used to make pixels react to specific colors of light. It seems strange to me that large areas tend to have tints when you average them out. This implies to me that there is some localized effect that is skewing the data, rather than a true random noise component. I'll look at multiple images tonight and see if the patterns seem fixed for a given camera. I'm pretty sure I checked this in the past, but it wouldn't hurt to revisit it and maybe get some examples to post.