Working with 14bit vs 16bit color in CS2 - does it matter?

[Deliberate1]

Sandy, thanks for bringing me to the answer.
And Ken, those are wonderful images you have. You have a good eye and the toning is exquisite.
David

[jetcode]

Some people argue there's no need for anything higher than even 8 bit. I don't believe it, but I do believe 14 bit will get you anywhere you need to go.

Also, you might want to re-check the specs on the eversmart. I've used one, can't remember what model it was, but from what I remember the scan was made at 14 bit and the down-rezed to 8 bit for the file. Don't know if they changed that on the II, but it would sure suck to get such an expensive piece of equipment and be dissapointed with it.

Again 14 bit means the measurement capability not the actual range of color expressed. Because 14 bits is available does not mean the data can be captured or expressed accurately. Sensors and print/display technologies have limitations. More resolution will buy more fidelity but at what point does the eye, the sensor, the printer, the screen limited the amount of fidelity that can be perceived. An example is the difference in audio.

8 bit: unusable
12 bit: voice grade quality (low fidelity)
16 bit: CD standard
18 bit: top fidelity
24 bit: no detectable difference from 18 bit

Part of this is marketing and part is technology. In measurement systems anything beyond 18 bits is noise, the natural chaos that exists in energy systems. In fact 24 bits is composed of a high number of samples averaged to reduce the noise floor. The ear cannot distinguish the difference.

For those who see these differences more power to you. I am not arguing 8 bit vs 16 bit. There is a difference. I am questioning accuracy and technological limitations.

Joe

[sanking]

Sandy, thanks for bringing me to the answer.
And Ken, those are wonderful images you have. You have a good eye and the toning is exquisite.
David

David,

Professional high end flatbeds are very heavy and cost a lot to ship. EverSmart scanners, for example, weight about 150 lbs and shipping half way across the country will run on average $300+ . And if the scanner arrives DOA you may be faced with very high costs to have the machine repaired. My advice is, 1) either buy from a dealer who will offer a guarantee that the scanner will arrive in good working condition, or second best, 2) buy locally from someone in your region so that you can physically inspect the scanner to make sure it is in operating condition, or 3) buy from someone who will offer you a money back guarantee if the scanner does not arrive in operational condition. And if you do #3, make sure that the amount of time you have to check out the machine is sufficient. Some of these re-sellers on ebay offer a 7-day return. It may take you three times that long just to set the machine up and get the software and everything else operational.

If you can not do one of the above you may wind up with an expensive piece of useless metal and glass furniture on your hand. I have purchased two high end scanners on ebay and both arrived in great physical condition, and with the status indicator lights working, but with problems that required a lot of attention and money to correct.

Buying a used high end flatbed is one of those times when it really is smart to think of the very worst case scenario. Same concern, or more, applies to drum scanners.

Just my two cents worth.

Sandy King

[Clueless Winddancing]

Re: Sand King's comment re "doing as much as you can in the prescan" when it is 14 Bit or less.
In the broad sense what would the workflow of that look like? Is it mostly dependent on the particular scanner SW used? Is it different for B/W? -for Reflection?

[Deliberate1]

Sandy,
I appreciate your advise. It is sound. I have been in the market seriously for about eight weeks. The choises are not so easy to make. You can buy and Eversmart Pro II from a refurbisher for $7500, or take a gamble on one on Ebay for $1000+/-. If it works, hooorah. If not, then you put $3-4k into it and get a truly refurbished machine with new parts that, hopefully, will last the life of the scanner. Or you can drop $8-10k on a factory rebuilt IQsmart2 or more for a 3 knowing the performance does not merit the price multiplier. Or a rebuilt Howtek 4500 directly from Aztek for the same. You took that gamble - twice. And I understand why. I read your posts. I would hope to get a Eversmart Pro II cheap enough, with the software, with the expectation that I will drop 3 or 4 times the cost on possible work. Or I could pitch the whole idea and just send my trannies out. Not likely for control freaks like me. To see those 6x6 trannies form the Rollie 6008/Nikon 9000 pop up on my monitor for the first time is bliss.That is the "gateway drug." Now I have bigger fish to scan....

[MJSfoto1956]

To the question; is there any significant disadvantage posed by a 14 bit file produced by Oxygen if I do similar adjustments in CS2? I suspect that I will be doing far more or the color correction, and so forth, in Oxygen which may eliminate any issues. That being said are there functions in CS2 that would likely degrade a 14 bit image in a way that I might not have experienced with my 16 bit Nikon?

With VERY FEW EXCEPTIONS, pretty much the best any CCD or CMOS scanner can do is 14bits of real data without supercooling. The last 2 bits are pretty much pure noise and most scanning software simply drop them (even though the file size is in fact, as others mentioned, 16bits)

So don't sweat it. 14bits *IS* 16bits for all practical purposes.

[Brian K]

It's not the 16.7 million bits that cause banding in an 8 bit rgb gradation, it's the fact that there are only 256 steps between pure white and pure black that makes the banding apparent. With higher bit depth you have smoother gradations and the difference between 14 bit (16,384) and 16 bit (65536) may rarely seem apparent in gradients, however every once in a while it might.


grayscale bit depth

8 256
9 512
10 1024
11 2048
12 4096
13 8192
14 16384
15 32768
16 65536

[Ken Lee]

While we are on the subject, can anyone give a layman's explanation of "wave"-based methods of data compression ?

Here is the extract of a US patent on one such approach:

"A method of image compression comprises the steps of recursively transforming a image using a Discrete Wavelet Transform. This creates a plurality of levels including at least a first level, multiple intermediate levels, and a low-low pass subband of the last level. The transformed image at each level is quantized, and datapacking the quantized image is performed. The step of datapacking further includes, encoding of the first level using adaptive run length of zero cofficients; encoding of the intermediate levels using run-length coding of zero coefficients and a predetermined two-knob huffman table for non-zero coefficients and encoding of the low-low pass subband using a low frequency packing algorithm".

[jetcode]

While we are on the subject, can anyone give a layman's explanation of "wave"-based methods of data compression ?

Here is the extract of a US patent on one such approach:

"A method of image compression comprises the steps of recursively transforming a image using a Discrete Wavelet Transform. This creates a plurality of levels including at least a first level, multiple intermediate levels, and a low-low pass subband of the last level. The transformed image at each level is quantized, and datapacking the quantized image is performed. The step of datapacking further includes, encoding of the first level using adaptive run length of zero cofficients; encoding of the intermediate levels using run-length coding of zero coefficients and a predetermined two-knob huffman table for non-zero coefficients and encoding of the low-low pass subband using a low frequency packing algorithm".

That description is in layman's terms.

The goal in all compression algorithms is to find patterns that can be reduced (encoded) and later expanded (decoded).

This compression technique sounds like (I have very minimal knowledge of compression techniques) it views an image as a linear series of bits, samples a given number of bits, and applies a transform to find frequency characteristics of the given sample stream encoding the results over a given series of frequencies (recursive). The encoded results are stored resulting in image compression. Note that the algorithm defines the way each level of encoding is computed.

Compression algorithms involve very advanced mathematics.

[jetcode]

It's not the 16.7 million discrete levels ...

As Brian points out the accuracy of color is specified by the resolution of a single channel not the combined potential of all 3 channels.