RAW Vs RGB - Please explain?

[Jonathan Burton]

The red workflow is very new to me and I am trying to understand it and wrap my head around it as best i can.

David Mullen ASC wrote this in a thread a while ago, and i was wondering if anyone would be able to explain clearly the difference between RAW and RGB, how it works and what it means to us in workflow:

"People here keep confusing RAW with RGB. 3K RAW and 1080P RGB are not all that different in resolution. Essentially 1080P RGB is 1.92K each for red, green, and blue, or about 6.6MP total. 3K RAW, if 3072 x 1728, is about 5.3MP.
You have to figure that 3K RAW will end up measuring around 2.5K hopefully when converted to RGB. That's not a huge leap over 1.92K.
Think of 3K RAW as being the "RAW" equivalent of 1080P RGB."


Any help would be great! Cheers.

 

[David Mullen ASC]

http://en.wikipedia.org/wiki/Demosaicing
http://en.wikipedia.org/wiki/Anti-aliasing_filter

A single sensor with a bayer color filter array has a mosaic pattern that filters 50% of the photosites green, 25% blue, and 25% red. From that pattern stored in a RAW file, a demosaicing or debayering algorithm has to create 100% green, 100% blue, and 100% red to create an RGB file. This causes some loss of resolution. Or to put it more bluntly, you're taking a single file and creating three files from it, one for each color. 3K RGB is three times the data of 3K RAW for example.

Combine that with the fact that all sensors have an Optical Low Pass Filter to roll-off high frequency detail in order to minimize aliasing.

So not only can't you resolve exactly the same number of lines in the subject as there are in the sensor because of the debayering process needed to create a color image from a RAW image, you actually might not want to due to aliasing anyway.

You have to think of a single sensor as creating a monochrome image, just one with a built-in mosaic pattern that has "clues" as to the color information -- later a three-color image has to be created from the RAW file using an algorithm that makes sophisticated guesses as to the actual color at each photosite location.

A RAW file is just what the sensor captured, after conversion from analog to digital, without any conversion step to become a color image.

You can also think of RAW as a very efficient way to record the image because it needs less data than RGB. In fact, three 1920 x 1080 files, one for each color, is slightly more data to record than a single 3K RAW file, if the compression rate is the same. If 3K is 3072 x 1728, for example, that's 5.3MP to record, whereas each 1920 x 1080 file is 2MP, so that's 6MP total to record for red, green, and blue.

 

[Jeff Kilgroe]

Edit> I thought I was fast, but David was faster. Haha.

This can get real complicated in a hurry and actual specifics will vary from one camera and/or sensor design to another. If you look at a full 1920x1080p raster, that is 2,073,600 pixels. Each having a component for Red, Green and Blue. So 1080p RGB has a maximum potential resolution of combined luma and chroma data of 6,220,800 or 3X the actual pixel count above. Whether or not a camera can actually produce that is another story. And you will find that most all high-end 1080p cameras out there that can produce a true 1080p full resolution raster do so by using much higher resolution sensors, multiple sensors and other techniques.

The RED (and most all CMOS based cameras) use a Bayer-Pattern sensor. So in the case of a 3K Bayer sensor, you get 3072x1728 pixels. That is 5,308,416 pixels. Half of which have an affinity for the green channel, the other half are split evenly between red and blue channels. So this means you get the full 5.3 megapixels of luma resolution, but your chroma resolution is quite a bit less. Effective or nominal resolution of a properly processed Bayer sensor is usually around 80%. So in this case, about 4.2 megapixels. Or what would equate to about 2.4K resolution. So 3K Bayer makes for plenty of resolution to do highly detailed and sharp 2K or 1080p images.

The difference between RAW and RGB is really how the data is recorded and at what point it is processed. RAW images are where the sensor data is recorded from the camera as unprocessed. You apply your color processing, debayer / demosaic algorithms, sharpening, etc.. in post when you "develop" your footage.

RGB footage is imagery that has already been processed through a debayer or demosaic algorithm, a color table applied, sharpening and scaling, etc.. In other words, its already processed into RGB. Whereas RAW, in the sense of REDCODE and their Bayer pattern CMOS, is recorded as a single 12bit luma value for each pixel on the sensor and each pixel only sees red, green or blue. Whereas a processed RGB image is that RAW data that has already been converted into a lower resolution (1080p) image that has full R, G and B color components assigned to each pixel.

The upcoming EPIC and Scarlet DSMC brains will record 1080p RGB (and very high quality) in addition to recording RAW. In most situations RAW will be superior, but the RGB record modes will offer some advantages such as faster turn-around times and less transcoding / debayer processing times in post.

 

[David Mullen ASC]

A lot this really goes back to an even more basic question, which is how do you capture color information of a subject? Some of the earliest full-color processes in still photography involved taking three separate b&w photos using red, blue, and green filters. In fact, that still works today (as long as nothing moves in the photograph.)

The first 3-color movie process was 3-strip Technicolor, which ran three b&w negatives through the camera. With color filtering, each b&w negative contained either the red, green, or blue picture information.

A positive was made of the negative with exaggerated densities so that it could be put in contact with either yellow, cyan, or magenta dye, absorbing the dye, and then that film (called a matrix) was run in contact with a clear strip of film that could receive the dye image. Then in three passes, you created a full color image on the print, similar to how color images are printed on paper. The printing process was called dye transfer, or imbibation (I.B.) printing.

Color negative film has very thin layers of silver halide grains separated by color filters so that each layer captures either the red, green, or blue information.

A 3-sensor video camera is a little like a 3-strip Technicolor camera -- the light is split by a prism, then passes through a filter before hitting one of three sensors. So again, each of the three sensors' output is monochrome but contains separate color information that can be combined later into a color image.

So to use a single sensor, you have to use a color filter array over the sensor (usually a Bayer pattern) that filters some individual photosites green, some blue, some red.

 

[Jonathan Burton]

Wow! Thanks for all the info David and Jeff, I really appreciate it!

Its one thing to know how to use a camera well and get the best out of it, but I am really trying to gain a reasonable understanding of how it actually works from shooting to post, and these forums have been so helpful!

 

[Jonathan Burton]

Just one more quick question to clarify, when people refer to .r3d files are these RAW?

 

[David Rasberry]

.r3d is Red's proprietary wavelet compressed raw file format.

 

[Jonathan Burton]

.r3d is Red's proprietary wavelet compressed raw file format.

Sorry, could you explain that. Are you saying .r3d is simply the file format that the RED RAW files are stored in, or are you saying these are files further compressed?:blink:

 

[David Mullen ASC]

Sorry, could you explain that. Are you saying .r3d is simply the file format that the RED RAW files are stored in, or are you saying these are files further compressed?:blink:

The RAW sensor capture is compressed using REDCODE and stored as a.r3d file. It's not further compressed.

 

[Jonathan Burton]

Got it! Cheers!