# Magic Lantern Forum

## General Discussion => General Chat => Topic started by: araucaria on March 01, 2014, 11:22:07 PM

Title: Naive question regarding bitdepth
Post by: araucaria on March 01, 2014, 11:22:07 PM
Please have mercy I have a very basic question.

When this raw craze started I was always calulating the speed requirements for certain resolutions so I knew which card I could use and which not.

X(pixels)*Y(pixels)*14(bit)*24(fps)/[8*2^20] would give me the Megabytes/second I would need. This actually worked.

I never thought much about it, but today I wondered that if it is 14bit/per sensel,shouldn't it be 4 times more data?

Any idea?
Title: Re: Naive question regarding bitdepth
Post by: g3gg0 on March 02, 2014, 12:29:03 AM
the formula is correct.

the bayer pattern has "less" resolution as you know it from RGB.
in RGB every pixel has three color channels, 8 bit each usually.

in bayer pattern every pixel is either red green or blue and during debayering in post processing,
you try to get the two missing colors for every pixel from those beneath.
(http://en.wikipedia.org/wiki/Bayer_filter)
Title: Re: Naive question regarding bitdepth
Post by: araucaria on March 02, 2014, 01:19:38 AM
Thanks for the answer, I actually had a completle wrong idea about how the bayer sensor worked, you made me read it again. Actually I thought every pixel was made of four single pixels... now I feel really stupid.
Title: Re: Naive question regarding bitdepth
Post by: ItsMeLenny on March 02, 2014, 01:43:09 AM
Thanks for the answer, I actually had a completle wrong idea about how the bayer sensor worked, you made me read it again. Actually I thought every pixel was made of four single pixels... now I feel really stupid.

Every resulting nonraw image is made of 4 bayer pixels merged into 1 pixel. A Red a Green a Green and a Blue.
Title: Re: Naive question regarding bitdepth
Post by: arrinkiiii on March 02, 2014, 01:47:32 AM

Talking about pixel and after read this... http://blog.planet5d.com/2014/02/is-dual-pixel-autofocus-coming-to-the-canon-eos-5d-mark-iii/   how can we tell if the sensor is up to this?
Title: Re: Naive question regarding bitdepth
Post by: araucaria on March 02, 2014, 02:04:59 AM
Every resulting nonraw image is made of 4 bayer pixels merged into 1 pixel. A Red a Green a Green and a Blue.
That's what I thought, but it's not.

Resulting image is 100x100pixels, then you have 100x100pixels on the sensor, the missing color information is borrowed from the neighbour pixels. I thought there were 400x400 pixels for a 100x100 image ^^'

This makes me wonder if there is any software where I could open a raw file and there will be no debayer guessing, just summing together 3 pixels to get RGB with the 4rth G to help reduce noise. Or is this just stupid?
Title: Re: Naive question regarding bitdepth
Post by: ItsMeLenny on March 02, 2014, 02:33:08 AM
Resulting image is 100x100pixels, then you have 100x100pixels on the sensor, the missing color information is borrowed from the neighbour pixels. I thought there were 400x400 pixels for a 100x100 image ^^'
I thought the same/ a similar thing at first. Then I thought "why doesn't it move one pixel every bayer pixel. Then I realized it does.

This makes me wonder if there is any software where I could open a raw file and there will be no debayer guessing, just summing together 3 pixels to get RGB with the 4rth G to help reduce noise. Or is this just stupid?
It's "just stupid". There's filtering techniques called bilinear and such, I guess in this case you would want "none" (I think it's usually referred to). But it's horrible, moire and aliasing. It basically does the opposite of what you would expect. The debayering technique you want (which most programs offer) is AMaZE, and the visual quality compared to linear is very visual, AMaZE has a nice crispness to it.
Title: Re: Naive question regarding bitdepth
Post by: dmilligan on March 02, 2014, 03:44:14 AM
It's usually called 'super pixel', there's no reason to ever use it really. Bilinear is pretty fast, VNG is a little better quality and AMaZE is about the best there is. There are also some proprietary algorithms (ACR).

I use a program called pixinsight that has various debayering methods you can choose from, including those mentioned above as well as a mode where each pixel is simply made the color of the sensel (basically you're seeing exactly what the sensor saw). That's very useful for comparison and analysis of raw data (but not for practical image processing). I guess technically that would be the most 'pure' way to display the data.