I don't believe any home desktop scanner does much work (image wise) in-the-scanner. Scanner sends raw data to software and the software does all the manipulation. And sharpness is certainly not pixel-based. How do you sharpen a single pixel?! You don't. You need surrounding pixels to build up local contrast and that creates greater perceived sharpness.
Otherwise, I do the same as you. I do all contrast and colour work before resizing. After resizing I do clarity and sharpness adjustments.
I'm going to assume you quickly skimmed my comment, then tossed this message back. It makes an absurd assertion; a picture with one pixel. Immediately after discussions of thousands of pixels per inch, I can only attribute to distraction or not enough time to work it through. Regardless way it is of no service to either of us, or othes for that matter.
I will admit, it was late when I quickly composed my message to you. If you look, it was after a much longer response to another member. Time is limited and in haste my choice of words was more expedient than fully descriptive.
I would appreciate you reconsidering what was posted to you, as well as additional information I provide next. If you still feel the same way after that, then I guess we can meet at dawn for 20 paces. LOL!
Some clarifications are in order:
And sharpness is certainly not pixel-based. How do you sharpen a single pixel?! You don't.
Technically all activities are pixel based, millions of pixels! In this circumstance the implied comment was "adjacent pixels", and only adjacent pixels. The algorithm is a 3x3, 9-pixel, matrix called a kernel. The center pixel is compared to the eight, surrounding and adjacent pixels, replaced by the result. Said kernel is offset by one pixel and the process is repeated for each and every pixel. I have written custom convolutions in the distant past.
This is what they look like:
-1 -1 -1
-1 5 -1
-1 -1 -1
I believe my intent was to *discourage* its use in scanner, if that was not conveyed well enough then this should remove doubt. There are better tools, such as unsharp masking (USM), for enhanced detail contrast. These tools allow adjustments to effectively detect period, the distance, contrast and amplitude of the original and apply it back to the image.
USM, for example, can replicate the effect of "pixel based" sharpening, such as the default of low-end scanning software. Many scanning applications do offer USM and it is easy to detect the difference. If there are no options, just "sharpen" and "sharpen more", for example, move along, there is little useful there.
USM is different from "pixel based" sharpening because it conceptually uses a second image. Traditionally it was done by hand, had been for many years before scanning was invented. It was a common technique in lithography to overcome the softening of the halftone process used in order to print. Even stochastic screening that has replace half-toning is a diffusion process, requiring USM to maintain perceived acuity.
You need surrounding pixels to build up local contrast and that creates greater perceived sharpness.
Please note the above does agree with you regarding how some forms of detail enhancement function. Hopefully this clarifies the misunderstanding and errors in communicaton. For further information, feel free to ask me, Professor Google, or Dr. Wikipedia. The Peach Pit book "Real World Photoshop" also has a reasonable description of sharpening and unsharp masking, at least the versions I edited were acceptable. I have no reason to expect a degradation of its usefulness.
I don't believe any home desktop scanner does much work (image wise) in-the-scanner. Scanner sends raw data to software and the software does all the manipulation.
The scanner you have on your desktop is substantially more capable than an optical mouse -- and an optical mouse is a scanner with a digital signal processor (DSP) embedded in it. The mouse does not push raw, pictoral, information to the host, only the interpreted motion data.
I don't believe any home desktop scanner does much work (image wise) in-the-scanner. Scanner sends raw data to software and the software does all the manipulation.
If your desktop scanner was not self-calibrating and correcting its linearity, you would be, you would have streaks down the scans, and like the rest of the scanner owners, in your shorts. ;-)
In other words, even when requesting "raw" data from the scanner, it is pre-processing the image. I made reference to lookup tables (LUT). Those are basic, necessary, tonal transformations.
Innumerable tools in image manipulation applications use lookup tables. The interface may use bezier curves and power curves (called gamma) to calculate the lookup values, then process. If you have Photoshop, call up the Curves dialog, click on the pencil and draw in the box if you don't believe me.
Without that rudimentary ability, even cheap scanners would not be able to quickly manage digital-to-digital (D-D) transforms. The cheapest of modern scanners are at least 10 bits A-D, requiring 10-bit to 8-bit conversions.
The scanner you have, Epson scanners and others, all allow the driver to load LUTs and bias the output. This is unnecessary if you are taking raw, 16-bit data (typically accurate to 12 or 13 bits, 14 for higher quality machines).
It is a different story when scanning into 8-bit per colour files (24-bit, RGB colour files). This is why I offered two options regarding scanning. What I didn't spell out was that doing so reduces the data to transfer, usually through pokey USB2. Though it is half the data, it does not cut the time by half.
Let me re-iterate that. Given an 11 or 12-bit accuracy scanner, scanning to 8-bit per colour files, the data to transfer is halved. This answers concerns of other members. In these circumstances, a scanner driver needs to load a gamma curve with end points (white and black levels) in order to bias the D-D conversion. Not perfect, but > 95% for most, normal images. The final tweaking will rarely be a quality issue.
All of that is moot if you choose to scan in 16-bit mode (48-bit colour).
Thus, my intent was to offer two options. As I said, other members had concerns about scanning at higher resolutions taking longer. There are lots of images that will reproduce well from 8-bit scans.
I will head argument off before it can rear its head. Regardless of my assertions, you need to know your own equipment. Take a good original and a calibration scale if you have one. Make a 24-bit scan without adjusting anything. Then, make a second scan with adjustments to white, black and gamma (curves are harder to replicate). Copy the first scan and apply the same controls, then compare the histograms of each (use a screen capture so you can put them side by each).
If the two adjusted images are identical, try a different driver. Also note, some scanners invert their curve notation. If this is the case, divide 1 by the gamma value. Therefore 1/1.8 = 0.555.
Otherwise, I do the same as you. I do all contrast and colour work before resizing. After resizing I do clarity and sharpness adjustments.
Actually, my workflow is broader than what was noted in that message but you had no way to know that. I wet mount my high-quality scans, for example, and have the luxury of some nice hardware to do it with.
I'm glad to hear you are doing some of the things I suggested at higher resolutions and bit-depths. It is apparent by the samples you posted that more is possible. The difference between the 1600 and 3200->1600 should be noticeable on many images, without degrading qualities.
If you wish to make available the 3200 dpi original of the 3200-1600 sample you posted, I will be happy to return it exposing the extra details available. Of course I will delete the file when I'm done, if you accept the offer. You will then have the option to post the results, I will not do that.
Additionally, "Clarity" is basically a combination of unsharp masking and colour saturation adjustments. Nothing wrong with the tool, just sometimes more or less of one is required. Knowledge is always key to solutions.
As well, you are sharpening upon previously sharpened image. Again, nothing wrong with that technique, if the values are different, I do it all the time, literally.
Finally, because 'clarity' has a colour-specific, not specific-colour, effect, it should be done at high resolutions as a general rule. In fact sharpening should be done at higher resolutions so the image looks appropriate at 25% zoom, 50% normal resolution.
Please ask for clarification if something doesn't seem right or if I've not explained well enough. Given the time, I'm happy to accommodate.
Scott
---
Please also note, my first response to this message is about double this length. I did describe supporting factors as well as alternate ways to understand USM, etc., but it was too much for one post. Even this is way too long, hopefully it is of use, I don't like wasting valuable time -- mine that of others.
