CMOS image sensor with adaptable histogram

On a sunny day (Sun, 7 Dec 2008 02:30:06 -0800 (PST)) it happened Jolly wrote in :

If I understand you right, then you would get a reduced contrast. The high brightnes parts would stop first, and later the low brightness parts. After some time all pixels would have the same high value...

If I understand correctly, this is like a seldom used method of photon counting, where each point or in this case pixel corresponds to the inverse of the time it takes to detect a fixed number of photons. If photon shot noise is the dominant noise source, this results in the same signal to noise ratio for each point/pixel. The contrast would not be reduced, but the exposure time for each pixel and hence changes due to changes in the scene would be different.

Bret Cannon

On a sunny day (Sun, 07 Dec 2008 17:29:55 GMT) it happened "Bret Cannon" wrote in :

So far I can follow you.

But here no longer. In a CCD the picture is read out at fixed time intervals, or at least after a fixed time, the exposure time. For a long enough time, something you want to get a usable signal, first the bright spots in the scene will reach the control voltage, then the less bright spots, etc. But after some time they _all_ have reached the control voltage. So then you read all whites, or all the same level.

Yes, but you cannot measure exposure time of a pixel or can you?

Hi Jan, hi Bret, thank you for your comments!

Your assumptions are right for the particular control voltage Vc(t) =3Dconst.; in this case one would have to measure the time until Vc is reached to get some information at all.

But what I had primarily in mind is the case of a really time- dependent Vc(t), e.g. a monotonically decreasing Vc(t). This has the following effects: Pixel number n builds up voltage like V_n=3DH_n*t, with H_n being its brightness. The buildup is stopped at an individual time t_n when a "final voltage" E_n=3D H_n=D7t_n =3D Vc(t_n) is reached. Th= e control voltage Vc(t) could then be adjusted such that the histogram w.r.t. the final voltages E_n is balanced, i.e. each possible final voltage between a maximum and a minimum is assumed by approximately the same number of pixels.

I put a pdf sketch of the idea at

On a sunny day (Sun, 7 Dec 2008 11:06:06 -0800 (PST)) it happened Jolly wrote in :

Interesting. It almost looks like you do the gamma correction in the sensor. For a gamma 1 ^1/2.2 ?

Hi Jan. I have only amateur experience in image processing, but as far as I can see from Wikipedia on the topic "gamma correction", you are right. With one addition: My Vc(t) could be quite arbitrary, e.g. also be discontinous. And it could be made dependent on the total current consumed for charge buildup by all pixels to achieve "automatically" an even distribution of final voltages E_n over the available range. But I do not know: 1) is it new? 2) is it realistic?

Regards, Jolly (Juergen)

On a sunny day (Sun, 7 Dec 2008 22:53:38 -0800 (PST)) it happened Jolly wrote in :

Normally in a video camera, as used for TV, there is a gamma adjustment control. So gamma is also variable (withing limits). If your idea is realistic, that depends on if you can get a CCD manufacturer interested to make one? You need to do some math to see if it actually brings some sort of advantage that is worth the huge investment to design a new type of sensor. I do not know if it is new, I have never seen this before, but that means very little ,as I do not follow this subject very closely, Technically I think it can be done.