photo sharing and upload picture albums photo forums search pictures popular photos photography help login
John Sheehy | all galleries >> Galleries >> demos > gradient.jpg
previous | next
06-FEB-2008

gradient.jpg

Demonstration of the fact that binning actually reduces highlight headroom.

A pure 0-255 gradient was made, then noise was applied, image was divided into 4 strips, from top to bottom:

Original Pixels
Pixels Binned 2x2
Pixels Binned 3x3
Pixels Binned 4x4


Then the image was clipped at 205.

The original, with its higher pixel-level noise, actually has more headroom.


other sizes: small medium original auto
comment | share
Samu-san24-May-2009 00:18
> Drawing that line further: could a gigipixel sensor where each sensel only hold a few electrons be the key to better highlight performance by having a real shoulder in the response, or is the effect neglibale nomatter how tiny the pixels are?

I simulated that hypothesis using poisson distribution for the photo-electrons, for 1-bit pixels, replacing a 32000 electron-well pixel with a sum of 2^15 1-bit counters (that could be the result of cutting the pixel in spatial slices, but also TIME slices if the read speed allowed it).
You can see the "shoulder" response in my post
http://forums.dpreview.com/forums/readflat.asp?forum=1018&message=31943093
jean paul 28-Oct-2008 11:22
I would think that large pixels do perform more 'accurate' in the highlights than small binned pixels if all other factors remain identical
In binned pixels the photon influx of the separate locations is averaged after registration while in large pixels it is averaged (spread out) during registration
When one of the binned pixels clips while the others remain below clipping, then the registered photon is still lost
In a large pixel, the photon is still correctly registered.
The 'better' performance of the smaller pixels is therfor due to the non linearity caused by shot-noise induced clipping
Drawing that line further: could a gigipixel sensor where each sensel only hold a few electrons be the key to better highlight performance by having a real shoulder in the response, or is the effect neglibale nomatter how tiny the pixels are?