Monday, July 20, 2009

Moore's Law and Camera Sensors

There is an interesting discussion on Luminous Landscape about Moore's Law and digital camera sensors. Moore's Law would suggest, e.g., that the number of pixels in a full-frame (35 mm) sensor should continue increasing to the hundred megapixel level; while the laws of optics suggest that such sensors would quickly resolve the tiny diffraction circles of even lenses at large apertures, and so increasing the resolution would be pointless.

It occurs to me that having lots of extra pixels might serve other purposes. For instance, it might be used to increase the dynamic range of the sensor. That is, the pixels on a CMOS sensor has a certain range of dimmest to brightest light it can register (a range of about 11-12 exposure values). At even brighter levels, the pixel simply responds at the top of its dynamic range; i.e., further increase in input light does not produce any (significant) change in what the pixel records.

Now, a very high pixel count sensor could devote two pixels instead of one to each picture pixel. (A picture pixel is already composed of R, G, B sensor pixels; we'd be doubling that). One of the pixels would be normal, the second would have an optical attenuating filter. So when the normal pixel is saturated, the filtered pixel would be somewhere in the middle of its range. It would be an easy process for the camera processor to pick the normal pixel, when the normal pixel is within its useful range, and to pick the optically filtered pixel when the normal pixel is saturated, and translate the output of the optically filtered pixel appropriately.

A similar effect could be had by having large pixels and small pixels. Say, a large pixel has twice the area of a small pixel. For a given exposure, large pixels would be used for the lower range of brightness and the small pixels would be used when the large pixels saturate.

This way, the very large sensor resolution could be traded for a smaller picture resolution but a higher dynamic range.

PS: I'm quite sure that camera designers have already thought of this. It might be interesting to dig out the relevant patents.

PPS: the small/large pixel idea won't work. A photographic scene is characterized by the light power per unit area of its parts; the choice of aperture and shutter speed reduces that from the point of view of the sensor, to light energy per unit area; and the same light energy per unit area would saturate large and small pixels just the same.


Anonymous said...

very intersting article. i also have an interesting article on this topic, i can fwd it to you if you leave me your email.

thanks for the great read!

banerjee said...


Post your article on Scribd if you're human.