Questioning 16 bits

[Drol]

OK, you have to read this carefully in order to understand it.

I scan 48 bit RGB picture.

I make two copies of same picture.
One pic is called "8.tif" other is called "16.tif"

I load 8.tif into photoshop, and convert it to 8 bit tiff and save it.

Now I have two exact images except "8.tif" has 8 bit color and "16.tif" has 16bit color saved.

Now I load both images into photoshop.

I make several, exactly the same, changes using curves and levels to both images.

After changes have been applied, I downsample 16 bit picture to 8 bit.

What I am expecting is that 8 bit image should be posterized and 16 bit is supposed to preserve smooth color transitions.

If I look at the histogram 8 bit image really shows color loss and one that was originally 16 bit shows no color loss on histogram.

BUT!!! -(now comes the interesting part)- BUT!!!

If I take one image and put it in a new layer exacctly over another, then change upper layers mixing mode to Difference i get a picture totally black.

To those that dont know what that means... it means 16 bit and 8 bit images are exactly the same.. there is no difference beteween 8 and 16 bit.

What causes that?

Maybe PhotoShop cant handle 16 bit data as it is supposed to?
Maybe SFs 16 bit per channel data exporter does something wrong?
Maybe Im doing something wrong?

Anyone got any idea?

You can try it yourself, because if this is the case as i described above, then SilverFast 16 bit option along Photoshop doesnt work...
... or maybe im doing something wrong?

Enlighten me.


V.

[President_LSI]

16 bit versus 8 bit compared

We will look into this effect more precisely, but so far your findings could be the result of the type of image that you have been using.

In order to significantly see the effect of 16 bit versus 8 bit processing, you could make a colour gradation in Photoshop, save one as a 48 bit tif file and the other converted to 8 bit and saved as well.
Load both into SilverFast HDR (one after the another). Then do those curve and other corrections, save as tif files again and compare the results as previously done!
This procedure should clearly show the difference!

[Drol]

We will look into this effect more precisely, but so far your findings could be the result of the type of image that you have been using.


Thank you

Vjr

[Drol]

Hello,

I had some free time yesterday so I made some tests.

My first theory was that SilverFast is working OK and Photoshop is handling 16 bit per channel (48 bit) images incorrectly.

I used my favorite 3D software and rendered an image in 48 bit.
It is a software that works with floating point colour number internally.. so basically it can achieve more than 16 bit colour images per channel.

I decided to save two RGB images. One as 8 bit per channel (24 bit) and other as 16 bit per channel (48 bit).

I load both pictures in photoshop. Extremely change curves on both pictures (exactly the same changes on each of em).

Any then after curve changes applied: 16 bit picture has a nice gradient, but 8 bit picture shows loss of color channels. And you can clearly see it.


Conclusion:
It must be some sort of problem SilverFast has with PowerLook 3000.
PowerLook 3000 is 14 bit per channel scanner. Maybe SF doesnt convert 14 to 16 bit correctly. Maybe that is a place where to start.


I can upload all those pictures if you want to see em: 3D image (8bit vs 16 bit) comparision and scan pic (8bit vs 16bit) comparision.


Vjr

[LSI_Belli]

I think that you already gave the answer to your question:

Your first mail tells that SilverFast Ai can scan in 16 bits mode without any problem and that Photoshop supplies a better result with the 16 bits image than with the 8 bits image.

Your second mail confirms that Photoshop works well.

So I assume that SilverFast and Photoshop can handle 16 bits images, but the difference function in PS lead you to think that something was wrong in SF or PS.

However, no sotware is free of bug and the problem might be specific to your type of scanner. So could you just check the following things:

1. how do you exactly do the difference between your two result images (8 and 16 bits) ? Because as far as the difference of channels is concerned, it seems to me that it only works with 8 bits images.

2. when you mean black? Does the difference image look black or is it really black, that means that the histogram is concentrated in the bin with value 0 (or the two bins 0 and 1 because of truncation)?

Best regards.

Thomas

[Drol]

Thnk you for reply Thomas... I will make a web page showing you what is happening.