RAW without interpolation.

[iDunno]

Is there a raw converter that doesn't interpolate when it demosaics?

Without interpolation, you'll have 1/4 of the number of pixels on the camera specs, but very high quality ones.

Diagram:



Each color pixel will consist of information from 4 sensor photosites.


You'd end up with a razor sharp image which only takes up 1/4th the size. Major file size reduction. Aliasing might be a problem, however.
Does this already exist? If it doesn't i'm going to start writing a script to do this.

[OsmosisStudios]

What on EARTH are you talking about?

[drr0b]

Hmm,.. aren't you just describing pixels? I don't THINK each camera pixel is a photosite; rather 4 photosites make each pixel. I think that must be the case, or 100% crops would look really stupid

[iDunno]

If you don't know how bayer interpolation works, look it up.
Bayer filter - Wikipedia, the free encyclopedia

I'm just asking if there is a raw processor which doesn't interpolate when it demosaics.

Originally Posted by drr0b

Hmm,.. aren't you just describing pixels? I don't THINK each camera pixel is a photosite; rather 4 photosites make each pixel. I think that must be the case, or 100% crops would look really stupid

You are partially correct. 4 photosites make one pixel,and then the camera interpolates that.
See the example picture:
http://en.wikipedia.org/wiki/Demosaicing

[OsmosisStudios]

In essence what you're asking is if there is a RAW converter that will allow you to edit at the photosite level and not at the pixel level? If that's the case, then there's only one system that sort of does something similar, and it's the latest Hasselblad backs which take 3 pictures per photosite and composite them. It doesnt let you edit at the photosite level, but it does give more lattitude for each photosite and therefore a more accurate pixel.

As far as computers (and software on them) are concerned, a pixel is the smallest unit of measure.

(As an aside, I find your tone to be VERY condescending, which is in no way working in your favour. I would strongly suggest you review your sentence structure when posting, because it's coming off as being very patronising)

[iDunno]

Originally Posted by OsmosisStudios

In essence what you're asking is if there is a RAW converter that will allow you to edit at the photosite level and not at the pixel level?

Um....I'm looking for a software that uses 4 photosites to make one pixel. All it needs to do is:

1. Read data from 4 photosites (R, G, G, and B) from a RAW file
2. Determine the Red, Green, and Blue values
3. Spit out an RGB pixel
4. And repeat.

Steps 1 and 2 are taken care of by DCRAW, and then I could make a script to do the rest.

Originally Posted by OsmosisStudios

(As an aside, I find your tone to be VERY condescending, which is in no way working in your favour. I would strongly suggest you review your sentence structure when posting, because it's coming off as being very patronising)

Oh haha sorry I was frustrated since I couldn't get dcraw to do what I wanted.

[OsmosisStudios]

Originally Posted by iDunno

Um....I'm looking for a software that uses 4 photosites to make one pixel. All it needs to do is:

1. Read data from 4 photosites (R, G, G, and B) from a RAW file
2. Determine the Red, Green, and Blue values
3. Spit out an RGB pixel
4. And repeat.

Steps 1 and 2 are taken care of by DCRAW, and then I could make a script to do the rest.

As far as I know, the RAW files don't contain photosite information, only pixel information, as described by your list there. Sensors/processors do this automatically. I could be wrong. Could you explain the benefit of using a photosite-level editor over a pixel-level editor?

[iDunno]

A 16 megapixel sensor has 16million photosites. Since bayer needs 4 photosites to make one color pixel, 16million photosites divided by four equals a 4mp image.

But wait! The average consumer won't possible understand that, so in order to have a 16mp image, the raw processor will enlarge the image 2x in each dimension, to end up with 16mp image.

That's why still images are blurry when viewed 1:1

If we skip the enlargement part, we would save alot of disk space! The image would be 1/4 in file size!

Okay, maybe I just have too much time on my hands since its spring break

Originally Posted by OsmosisStudios

As far as I know, the RAW files don't contain photosite information, only pixel information, as described by your list there. Sensors/processors do this automatically. I could be wrong. Could you explain the benefit of using a photosite-level editor over a pixel-level editor?

Actually, they do, since RAW = sensor data before processing. A program like dcraw can convert each photosite's data to a grayscale pixel of it's value, using the command "dcraw.exe -d" I don't know how it converts each RGB value to 8bit though.

There are some people who managed to remove the bayer filter, thus making photosites non-color specific, which makes a non-bayer (and therefore non interpolating) grayscale sensor. Then, they shoot in raw and use dcraw with the command above to turn the values of each photosite into a pixel.

So a 16megapixel sensor modified this way would create a true 16megapixel (grayscale) image, instead of a 4megapixel (color) image which is then interpolated to 16mp

[drr0b]

Originally Posted by iDunno

But wait! The average consumer won't possible understand that, so in order to have a 16mp image, the raw processor will enlarge the image 2x in each dimension, to end up with 16mp image.

That's why still images are blurry when viewed 1:1

If we skip the enlargement part, we would save alot of disk space! The image would be 1/4 in file size!

OK, I see where you've gone wrong. You don't just group the photosites in fours to make one pixel; doing that would throw away most of the information. Let's say that the raw pixels are laid out as follows:

R1 G1 R2 G2
g1 B1 g2 B2

You are saying that pixel 1 is made of R1, G1, g1 and B1, while pixel 2 is made of R2,G2,g2, and B2.
Instead, pixel2 is made of G1, R2, B1 and g2, while PIXEL 3 is made of R2,G2,g2 and B2. Hence, you end up with 16 million pixels all with different colours. (You can get 16 million greyscale pixels too).
I hope this helps

As for those people getting individual grayscale photosite info, they're doing a nice intellectual excercise, but not improving their photos

[iDunno]

Originally Posted by drr0b

OK, I see where you've gone wrong. You don't just group the photosites in fours to make one pixel; doing that would throw away most of the information. Let's say that the raw pixels are laid out as follows:

R1 G1 R2 G2
g1 B1 g2 B2

You are saying that pixel 1 is made of R1, G1, g1 and B1, while pixel 2 is made of R2,G2,g2, and B2.
Instead, pixel2 is made of G1, R2, B1 and g2, while PIXEL 3 is made of R2,G2,g2 and B2. Hence, you end up with 16 million pixels all with different colours. (You can get 16 million greyscale pixels too).
I hope this helps

Ah! If I understand correctly, it's something like this?

Thats genius ! It'll end up with an 8megapixel image, right?
Which is then resized to 16 megapixels?