There are some special and curious cases, when you use a Bayer sensor, where the approach taken on my editorial is slightly different. Think that a sensor could somehow take a "contact proof" of a given pattern. Pixel pitch is not taken into account for these cases. I will consider no noise and a full-frame sensor to keep the "contact proof" idea.
A Bayer sensor as you might know has the following pattern:

Case 1

Let us apply the following contact proof into the above sensor:

As each Bayer reading needs 4 pixels (2 x 2) we will have for the top left hand corner:

G=0, R=1, B=0, G=0 total reading R=1

Shifting one pixel to the right we have:

R=1, G=0, G=0, B=0 total reading R=1

You can carry on like this for 16 million more readings if you want. The final result will be:

Don't panic. This is actually very good. 50% of the readings are accurate which is even better than what I mentioned on the editorial. You could say that this never happens and yes it doesn't. The output of a Bayer sensor is pretty much what I mentioned. You can also say that if you shifted the pattern by 50% to the left you would get a better result so lets do it!

G=0, R=0.5, B=0, G=0 total reading R=0.5
and
R=0.5, G=0, G=0, B=0 total reading R=0.5

You would end up with this:

As you can now 100% of the readings are wrong! This is an image full of noise although you cannot see it. It has nothing to do with the original information.

Case 2

Let's use now a "contact proof" of a Bayer pattern on top of the Bayer sensor. You will have:

G=1, R=1, B=1, G=1, all =1 white
and
R=1, G=1, G=1, B=1, all=1 white

As you can see now 100% of the readings are wrong! This is an image full of noise although you cannot see it. It has nothing to do with the original information.
Can we ever have a 100% accurate readings with a Bayer sensor? Yes!

Case 3

G=0, R=1, B=0, G=0 total reading R=1
and
 R=1, G=0, G=0, B=0 total reading R=1

which gives this:

As you can see now 100% of the readings are right! However there is a catch. There is no information whatsoever (well just a tiny bit) to be reproduced by the sensor. You would not need 16 megapixels to do this (or even 8 or 2). The complexity of this signal is infinitesimal.
You can do the same for full framed one coloured "images". Bayer always works and reproduces the original signal in its entirety, 100%. Unfortunately, unless you want to photograph A4 blank pages, this has no interest whatsoever.

back to editorial

© 2006 Paulo Ferreira - All rights reserved