Bit depth myth?

[davidh]

I sort of believed that a higher bit depth better might prevent image degradation during post processing. I am sure I read this in more then one articles. So posterizing brought me to the following test:

example one:
Step one: I converted an image to 24 bit and applied an aggressive brightness curve: hist 1
Step two: The same brightness curve was applied to the result of the step one: hist 2
Step three: The result of the step one was converted to 48 bit (no change in histogram): hist 3
Step four: The same brightness curve was applied to the result of the step three: hist 4

Example two (lightened to better distinguish gray):
the same steps except on a posterized image.

In the example one the hist 2 surprisingly came out better in mid tones then hist 4 (which came out a bit better in shadows) though their starting histograms looked equal. The one step more - conversion to 48 bit - seems to have caused the degradation, unless there is something much more complicated behind the scenes.

Something similar happened in the example two. Example two was made just out of curiosity what will happen to those columns.

David

[jsachs]

Sorry, I re-read your post several times and still have no idea what the histograms represent.

[Charles2]

There is probably an error in labeling the histograms and in procedure. Hist 2 looks identical to hist 1. Hist 3 does not look identical to hist 1.

[davidh]

Charles, of course you are right, the labeling is wrong. Hist 3 and hist 2 should be swapped.

Here it is once more:
image one:
24 bit input image -> applied brightness curve (hist 1) -> applied brightness curve (hist 3)
image two:
24 bit input image -> applied brightness curve -> converted to 48 bit (hist 2) -> applied brightness curve (hist 4)

The bottom line is that though hist 1 (24 bit image) and hist 2 (48 bit image) look the same, the second brightness curve application - has stronger impact on the 48 bit image than on the 24 bit image while I would expect it to be the other way round . It seems that the conversion to 48 bit has an adverse impact on the image where I would expect none.

Sorry for the confusion.

David

[jsachs]

I'm not sure how you generated the images and histograms, but when I went to reproduce the problem I discovered a fairly serious bug in Brightness Curve that caused incorrect results when selecting a 48-bit image into the Input Image control where a 24-bit color image had been previously selected. This bug will be fixed in the next release. After I fixed it I did not see any significant difference in the histograms of the same curve applied to 24- and 48-bit versions of the same image.

[davidh]

I just applied twice a curve similar to the attached one to a 24 bit 2868 x 4320 tif test color image. The histogram of each step was read from the Histogram tool. Then I repeated the procedure to the copy of the image that had been posterized down to four colors.

David

[jsachs]

Yes, but how did you apply the same curve to both images.

Did you

1) Create the curve, click Apply for the first image, then click on the Input Image control to switch input images and click Apply or OK to create the second output image

or

2) Load a saved curve and run two separate instances of the Brightness Curve transformation.

or some other method?

If you did 1), then there was a bug that I just fixed that would have given incorrect results.

[davidh]

I used the second method with reopening the BC dialog box but without saving and reloading the curve since the its definition is remembered.

Anyway, if the method 2) I used is OK, why has the conversion to 48bit between two applications of BC such an adverse impact on the result - see the attachment ?
I then tried on the same test with the curve saved and reloaded and other combinations of both methods but always with the same result. Is it a bug or should the conversion from 24 bit to 48 be avoided?

David

[jsachs]

Based on the workflow you are describing, you may be misinterpreting comments about bit depth. If you start with a 24-bit image, converting to 48-bit does not add any information to the image or confer any benefit when applying curves. The benefit of 48-bit images comes mainly if you start with a 48-bit image such as you might get from a raw camera file or a scanner. Such an image does have information beyond what is in a 24-bit image and that information can help reduce posterization when applying curves, notably where the slope of the curve is steep. Once you are done making tonal modifications, you lose very little by converting at that point to 24-bit.

There are some very minor benefits to converting a 24-bit image to 48-bit in that sharpening and blurring may be slightly smoother, but it is unlikely you could see the difference by eye. Certain color space conversions may also be a little less likely to posterize in 48-bit color, but again the effect should be very minor in most cases. Personally, if I have a 48-bit image I tend to leave it as 48-bit and if I have a 24-bit image I leave it as 24-bit.

The difference in histograms you are seeing is apparently due to a change I made in PWP 7.0 in the way 8-bit images are converted to 16-bit images (and 24-bit to 48-bit) in that a small amount of random noise is added to each channel during the conversion (this is called dithering) which makes for a slightly smoother looking image. In earlier versions of PWP, it the operation you describe in your post produces more or less identical histograms. The bumpiness of the histogram actually indicates less posterization in this case.

[davidh]

I usually convert to 48-bit only after posterize transformation which produces 24-bit result. I have always done it not to improve the result but to lessen possible degradation of the image if it needed additional contrast change or sharpening. Now it seems using convert in this procedure is redundant and not really worth it so I may as well leave it to 24-bit.
The same will be true for Watercolor transformation as it also produces 24-bit images.

Jonathan, thanks for clearing it up to me.

David