Earthbound Light - Nature Photography from the Pacific Northwest and beyond by Bob Johnson
Home
About
Portfolio
Online Ordering
Contact
Comments
Recent Updates
Support

Photo Tip of the Week
CurrentArchivesSubscribeSearch

Are 14-bit Raw Images Really Any Better Than 12-bit Raw?

Just when you think you've gotten a handle on the JPEG versus RAW debate and which color space is better than the other, along comes a new choice to make. From the earliest days of support for raw file capture with digital cameras, they have always been 12-bits per pixel. Now both the Nikon D300 and D3, as well as the Canon 40D can do 14-bit raw captures. But since 14-bit files are obviously bigger, are they really any better?

By now, many people have done at least some degree of testing in an attempt to answer this question. In side by side comparisons, most reviewers have been unable to find any difference between 12-bit and 14-bit raw. I must admit that I initially concluded this as well based on shots taken with my new Nikon D300 but was still puzzled as to why. I mean, they included this in the camera for a reason, right? Some people have suggested that the difference was there even if your computer monitor and your printer were not up to the task of showing it to you. But this doesn't strike me as a very satisfactory answer. There has to be more to 14-bit raw capture than mere marketing hype.

To figure this out, let's first consider what a raw file actually is. Unlike a normal image in an RGB color space, a raw file only has a single channel of data. Individual photosites have a red, green or blue colored filter over them to allow only light of a particular color to register, and the color of adjacent photosites differs so that collectively they form what is known as a Bayer mosaic pattern. Thus, at each photosite, only a single color is recorded (red, green or blue rather than all three), and the values for the two that are not present are later filled in through a clever process known as demosaicing.

But even discounting color, the data that each photosite registers differs radically from how we actually see the world. Human vision sees doubling and halving of brightness as being jumps of equal magnitude, a concept that correlates with the way photography measures increasing and decreasing brightness in terms of "stops." The values recorded by each photosite though are essentially simple counts of the number of photons that strike each. If one assumes there exists a maximum value that can be recorded by a photosite (a safe assumption since we all know what a burned out, overexposed image looks like), this means that if we record precisely half that value we are only one stop below being at pure white overexposure. Cutting this value in half again takes us down another stop, and so on. Thus, while one might assume that equal changes of brightness increase the photosite data value by equal amounts, they actually must double or halve it. This is known as "linear gamma" as opposed to the 2.2 gamma generally used to create computer monitor displays that match more closely to the way we see.

Now we are ready to consider how changing the values recorded in a raw file from having 12 bits to 14 bits might affect things. They can't provide for recording a wider range of colors or a greater dynamic range since these are a consequence of physical properties of the sensor itself. Rather, the extra bits provide for greater precision with which each photosite value turns into a recorded number in the raw file.

A binary number with 12 bits of precision can record a number with 4096 different possible values (2 to the 12th power). Jumping to 14 bits gives us 16,384 possible values, or four times as many as with 12 bits. Successively dividing these maximum values in half gives us this many values for each stop:

   12-bit   14-bit 
Brightest stop 2,048 8,192
Second brightest stop 1,024 4,096
Third stop 512 2,048
Fourth stop 256 1,024
Fifth stop 128 512
Everything else 128 512
Total distinct values 4,096 16,384
Test chart shot at four stops underexposed
Test chart four stops underexposed

It's because of this that shadow detail has always been a problem in digital photography. Based on the table above left you can see that the number of values available for shadow detail in 14-bit mode matches that of 12-bit mode two stops brighter. It would seem then that this might be one area where having the extra bits might help, and sure enough my testing has confirmed this.

Shooting a lens test chart four stops under exposed is perhaps not something most people aspire to in their photography, but in an effort to test for shadow detail, this is exactly what I did. You can see the almost black frame on the right above next to the chart of stops. I shot this in both 12-bit and 14-bit raw on the D300 and then post processed it using both Nikon Capture NX 1.3 and Adobe Camera Raw 4.3.1. The results from each were similar though not identical. I was able to get better results from ACR though perhaps because NX only supports a change of up to the first two stops exposure compensation during raw conversion whereas ACR allows up to four stops. The rest of the exposure increase in NX had to be done with D-Lighting and Levels.

Shadow detail recovery in 12-bit versus 14-bit
Shadow detail recovery in 12-bit versus 14-bit

One of the first things I found was just how remarkably good the camera did in both 12-bit and 14-bit. I first tested with only modest underexposure but found that unless I pushed it as far as I did both bit depths produced virtually equally exceptional results. Starting from a full four stops under though I could detect a difference in the finest details captured. Shown here are 200% enlargements taken from each ACR-processed image, and I have marked the area these samples were taken from on the full frame image. As you can see, the area is all but solid black yet I was able to pull quite a bit of detail from both images.

Test chart shot at four stops overexposed
Test chart four stops overexposed

Test chart shot exposed normally
Here's the test chart at normal exposure

While there are obvious differences at the coarser (left) end of the chart, the finer details on the right end benefit greatly from the extra data that 14-bits provide. With so few bits available at the deep shadow end in 12-bit, everything blurs to a fuzzy gray while having more possible values at that level in 14-bit mean the sensor was able to more accurately record when the bar should be black and when white. For extreme shadow detail recovery, 14-bit raw is a winner.

The same doesn't really hold true though in the case of highlight detail recovery. At the shadow end, you can keep cutting the remaining values in half to go ever darker into the murky depths so more possible values mean you can go further down into the shadows and still have enough precision to be useful. At the highlight end, you have excellent detail right up to the point where you go over the highest value the sensor can record. Anything from 4096 on up in 12-bit mode or from 16,384 in 14-bit renders as pure snowy white with no detail at all. You simply fall off the cliff as it were. Whereas shadow detail recovery works primarily by pulling the extremely low values captured in the deep shadows up into the range of useful brightness, highlight recovery can't pull details from beyond the top of what the sensor can record. It simply isn't there to be recovered. Instead, highlight recovery for one channel consists of a further round of interpolation based on what was captured in other channels. That is, if red is burned out beyond the top of possible values, you may be able to guess some of the details from what got recorded in the green and blue channels. So long as at least one channel has useful detail in it, the raw converter can use it to attempt to reconstruct what is missing from the others. If you burn out all three channels, everything is lost at that part of the image, but so long as the raw converter can find some useful information at that or neighboring points in other channels it will do its best to figure things out. So while the extra precision of 14-bit may help some, we already have a good number of available values in the brightest stop in 12-bit. 14-bit doesn't have that much it can bring to the table here.

To test this, I shot the same test chart at four stops over and got an image bordering on pure white. Working from the same part of the chart, I was able to use highlight recovery in both ACR and Capture NX to get quite useful results from both 12-bit and 14-bit captures. Yes, the 14-bit may win by a hair, but both are quite good.

Highlight detail recovery in 12-bit versus 14-bit
Highlight detail recovery in 12-bit versus 14-bit

You may wonder how a raw converter was able to pull out any detail at all if the test chart was close to pure white to start with and then overexposed to boot, but remember that it has to guess that it is pure white in the first place. In the middle of a white area, a red photosite and its adjacent green and blue photosites will all record a high value, but at an edge between white and black, green and blue photosites over a black area my record low numbers while a neighboring red photosite records a high value. The raw converter must interpolate where the exact edge is between black and white. In areas of extremely fine resolution, detail will begin to blur to gray. As long as any differentiation between black and white still exists though, highlight recovery can do its thing. But it does it nearly as well with 12-bit as it does with 14 because values within the brightest stop are already fairly precise even in 12 bit.

By now you're probably wondering whether I'd recommend shooting in 14-bit mode since the benefits seem to be modest for correctly exposed images and only seem to become significant at extreme underexposure. Right now, I'm not sure. When I first got the D300 I went through all the menus setting them the way I thought I wanted them and I set raw to 14-bit capture mode just on principle. More is always better, right?

14-bit captures end up being only about twenty-five percent larger than 12-bit files, but the camera slows down considerably to get them. Maximum frame rate on the D300 drops from 6 frames per second with 12-bit raw to only 2.5 frames per second in 14-bit mode. For landscape work, it's probably acceptable, but for wildlife or sports shooters this would be painful. You'll probably need to consider what you are shooting and what the lighting conditions are like in order to decide but it would seem that 14-bit capture is not appropriate for everything. For now, I'm leaving my D300 set on 14-bit, but I have added the raw recording mode to the new "My Menu" Nikon thoughtfully provided. I suppose they must have known I would need to change back and forth even I hadn't realized until now.


Date posted: January 13, 2008 (updated January 14, 2008)

 

Copyright © 2008 Bob Johnson, Earthbound Light - all rights reserved.
Permanent link for this article
 

Previous tip: Gorillapod: Searching for a Good, Small Outdoor Tripod Return to archives menu Next tip: 14-bit Raw Versus 12-bit Raw: Further Analysis and Comparison

Related articles:
Just What is a RAW File Anyway?
First Thoughts on Nikon's Big Announcement
 

Tweet this page       Bookmark and Share       Subscribe on Facebook via NetworkedBlogs       Printer Friendly Version

Machine translation:   Español   |   Deutsch   |   Français   |   Italiano   |   Português


A new photo tip is posted each Sunday, so please check back regularly.


Support Earthbound Light by buying from B&H Photo
  Buy a good book
Click here for book recommendations
Support Earthbound Light
  Or say thanks the easy way with PayPal if you prefer



Home  |  About  |  Portfolio  |  WebStore  |  PhotoTips  |  Contact  |  Comments  |  Updates  |  Support
Nature Photography from the Pacific Northwest and beyond by Bob Johnson


View Cart  |  Store Policies  |  Terms of Use  |  Your Privacy