Home

Robert D Feinman

Is 16 Bit Editing Necessary?



There has been much discussion of the need for 16 bit color depth when capturing and editing files. This tip will try to highlight the differences.

First let's discuss the theory. The graph at right shows what happens to the density of the image as we progress from a bit value of 1 up to higher numbers. So from a value of 1 to 2 the density changes by 50%, but from 7 to 8 the change is only about 2%. Thus round off errors which occur during image capture or editing will only be large for the darkest portions of the image. This is also the region in which the eye has the least ability to discriminate small changes in brightness. In the following steps we treat the same image with a bit depth of 8 and 16.

Density steps
Percentage Change in Density
vs Bit Value

This is the full frame of the image we will work with. The area that I was interested was the person in the red outfit in the center of the frame.
Unfortunately I underexposed since the person is in shade, and has on a dark colored outfit.
This image will require a large editing change to bring the colors out properly.

The other thing to keep in mind as we progress is the smallness of the areas we will be concentrating on.
The image was photographed on 200 speed color negative film and scanned in at 5400 dpi with the Minolta Dimage Elite 5400 scanner.

Full Frame
Full Frame Image

Here is a section that we will edit in the following steps. Even though I wanted just the person, I have left some of the sky in to show what happens after editing. Two version of this image were created. One scanned at 8 bit depth and the other at 16 bit. Visually they look the same at this point.

Part of Interest
Segment of Interest
These are the histograms of the two versions of the image.

Even though the images appear the same on the screen we see that there are differences in the files.
 
The 8 bit file is more spiky than the 16 bit. This indicates that the tonal values are segregated into a smaller number of "bins".
16 bit histogram
16 Bit Histogram
8 bit histogram
8 Bit Histogram
We apply this curve to the image to brighten the foreground. The dark values which had a minimum of about 11 have been pushed to 0. The highlights, such as the interior watch box lids in the lower left of the picture have been moved to about 250. Thus we have used only about half of the values present in the original image. These have been spread out over the 0 to 100% range of brightness.

Correction Curve
Correction Curve

Here is the same area after the adjustments. The foreground and the person in red are much better. The sky is completely blocked up, but in the real version we cropped this out.
After Correction
  After Correction Curve

Here are the two histograms again.

Notice that the range of brightnesses is better distributed from dark to light.

In the original the values were bunched up at the left side due to the underexposure.

Also notice that the 8 bit histogram shows a severe "comb" effect. This is caused by rounding errors as the small number of original density values are spread over a larger number of "bins".

16 bit histogram after
  16Bit Histogram- Corrected Image
8 bit after correction
8 Bit Histogram - Corrected Image
These two images are screen shots of the images zoomed in to 300%.
To my eyes the grain detail seems a little more distinct in the 16 bit version and perhaps the stars are slightly better defined.

To maintain a sense of proportion refer back to the full frame image to see what a small region we are dealing with.

We are showing screen shots here since there is no way to output a version of this image at such a high magnification without introducing artifacts from the jpeg compression process.
16 bit magnified
16 bit at 300%
8 bit after
8 bit at 300%
Here are portions of the actual files saved as minimal compression jpegs.

If we displayed the full frame image as an inkjet print at this size it would print out as about 18x24 inches.

Can you see a difference?

The following panel shows what has happened.
16 bit 100%
16 bit 100%
8 bit 100%
8 bit 100%
This is a view of the differences between the two images.
They were superimposed, the blend mode was set to difference, and a very steep curve was applied so that the changes would be emphasized.
Remember from the original discussion that rounding is only going to cause a 1 step change in value and that the darkest tones will be effected the most.
The answer to the original question is that there are differences between 8 bit and 16 bit editing. They can be seen with careful scrutiny, but are awfully hard to notice when images are presented normally.
So, should you use 8 bit or 16 bit?
Yes.
differences
Differences
Sponsored Advertisement


Back to Tips

Take some time to enjoy the rest of the site
Send email to robert.feinman@gmail.com

© 2003 Robert D Feinman