3 ms·
> I'd imagine the dynamic range sucks though if each pixel is so small that it saturates quickly. Well, if you replace a single sensor with four sensors that s
by klodolph 15y ago
> I'd imagine the dynamic range sucks though if each pixel is so small that it saturates quickly.
Well, if you replace a single sensor with four sensors that saturate more easily, observe that the chance that all four sensors saturate may be less than the chance that the original, larger sensor saturates. Answering the question, "how much light actually fell on these four sensors" becomes a more difficult problem and may yield greater error but is still tractable.
For you with a math background:
Let X_i be the amount of light that actually falls on sensor i, for i=1..4, and let e_i be the amount of uncorrelated noise. The sensor readouts are Y_i, where Y_i = min(X_i + e_i, sat) and sat is the saturation. Let X=sum(X_i). We know that X and each X_i follow approximately a Poisson distribution and we wish to know the parameter of the distribution for X.
Obviously, if Y_i = sat for all i then we are SOL. However, if not, then we can always use some kind of reasonable estimator for the desired parameter.
- Gravityloss 15y agoWell, say you take a back-lighted image of a person against the sky. Your camera software chooses the shutter speed, aperture and iso sensitivity according to some formula like average lighting in the scene. But the real world has such huge dynamics so that basically either the whole sky is washed out or then you can't make out the person's face. So basically all the pixels where the sky light falls on, oversaturate. If you lower the shutter speed, sensitivity and aperture so that the sky stays just below saturation, then you conversely catch very very few photons from the person. The person may appear pitch black or at best, really noisy, since the photons are so few. With noise, you often actually get color noise since because of the Bayer mask, the single photons hit random color pixels here and there. It looks really bad and cheap. Actually, that's the tell tale sign of low dynamic range and tiny pixels. What do people do with even 12 megapixel 4000x3000 images? I'd settle with 2000x1500 - still greater than any resolution on any of the screens I own, and the dynamic range could potentially be better. The full res images look really bad and noisy anyway. Actually, most of internet's picture content is just too large photos which for the most part don't contain any useful information: below some resolution you are just recording noise. I've set my compact to 5 megapixels. Yet we haven't seen the end of wasted bit proportion. Phones recording full HD video and later 4K and 8K will explode the data demands a thousand times. Everybody will watch those resized to much smaller resolutions too.