High Quality Scanning

[Piupianissimo]

We all know, that IMSLP is the biggest source in the internet for PD music scores. And every day the number of scores is growing rapidly due to the big number of contributors.
But only part of the scores are in good printing quality. So when I started to contribute my aim was to offer High Quality Scans and I hoped, that I could inspire others to try the same.
Now some members asked me about the technique I use for scanning and cleaning scores. Therefor I wrote a little cookbook you can find here.
As I kept it rather short, I invite everybody interested to ask questions so I can add additional information needed to make use of it. Perhaps you have other ideas or techniques, know better software or filters, that can improve the pages content .

[horndude77]

15-30 minutes per page seems excessive. This should be minimized if one ever wants to scan a lot of pages. Daphnis is the most prolific scanner here. I wonder what minutes per page he usually achieves?

Lately I've been scanning (flatbed) to 600 (or 300) dpi grayscale, then automatically post-process that to 600 dpi b&w (deskew, automatic thresholding, some noise removal). It gives me good results. I haven't timed myself, but I'd bet I get about 4-8 minutes per page. This is still slow but a lot of the time is in the scanning itself and almost all of the post-processing is automatic. Ideally I'd have an overhead stereo scanner with automatic dewarping (I can dream).

[vinteuil]

I added it as an aside to the Contributor Portal - you can find it under "Scanning Scores" or whatever its called

[Piupianissimo]

Hi horndude77,
you are right, 15-30 minutes per page is much time. Therefor I didn't contribute more scores to IMSLP now, althou I own some hundreds of good old scores.
Perhaps one should look at the total time, dividing it into the single steps needed:
1. cleaning the scanner surface and putting the score onto the scanner takes about 30 soconds
2. the scanning process itself depends strongly on the scanner hardware - in my case it's between 1 and 2 minutes for a 1200dpi scan. In the worst case the score wasn't good enough adjusted, so I have to rotate it a bit and redo the scan.
3. the filtering with the software takes about 1-2 minutes - again a hardware dependence
4. the manual cleaning takes 5-10 minutes (any pencil remarks to remove?) - I do that in 50% zoom and the 1200dpi score is a big picture to scroll thru
5. adjusting pagesize and borders takes 1-2 minutes for me
Some of the steps can get paralelized, you can work the filter while scanning another page or clean a page manually.
All in all the biggest amount of time is the manual cleaning process. If you have real good score quality you may perhaps want to do without this step. This saves you 2/3 of the whole time.
The main advance you get out of the 1200dpi scanning, this gives the filter more points to work with and you won't loose to much information thru the filtering process in my opinion. So if you own fine fast running hardware and paralelize the working steps and do without the manual cleaning you may perhaps end up with high quality scans that take only 4 minutes per page total working time.
At least me, I don't like any low quality printouts, so getting the scans fast is always a good thing but not at the expense of low quality (whatever is "high" and "low" quality). Let's try to get the scans indistinguishable from the real thing and call this "High Quality Scans".

[Piupianissimo]

Hi perlnerd666,
thanks for you adding the links to the page. I'll try to go on with my efford convincing contributors from the "quality way" (the "cookbook" can still get a lot better, I know). BTW: I wish there was a search option listing high quality scans available on IMSLP, so I could easily find them.

[vinteuil]

Jut make a category (Category:High Quality Scans, maybe), and then include [[Category:Category Name]] in the Misc. Notes section of each contribution. Feel free to add it to any scans.

[daphnis]

While I will say some of your work is very very good, I find this high quality method to be WAY overkill...for music scores. Remember we're dealing with some of the simplest image complexity possible, so unless you have some very specific requirements, there's no need at all to go over 600dpi. 300-600 is just fine for everything from viewing to printing, and I'm a daily practicing musician who very often, I dare say daily, uses his own scans for both viewing, and performing from printed versions. I never have problems with viewable/printable quality, and with my workflow I spend about 5 seconds per page scanning (at max. double that to include manual post-processing).

[Piupianissimo]

..there's no need at all to go over 600dpi. 300-600 is just fine for everything from viewing to printing...

I share your view - althou I would say you can really see a difference between 300 and 600dpi when you print the score (300dpi may still be enough to play the score). My intension is that the quality is good enough to reproduce the score pefectly when printing it. And therefor I'm convinced 600dpi are needed. The reason why I'm still scanning with 1200dpi is, that the filter I use needs at least more than one pixels to work propper in cleaning the scan from dust pixels. Now, when I work the filter on a 600dpi scan I loose to much information of the original score (e.g. some notes tails get lost).

...with my workflow I spend about 5 seconds per page scanning (at max. double that to include manual post-processing)...

I wonder how you manage a manual postprocessing of the scan in only a few seconds. As the 600dpi (or even 1200dpi) scan is rather big I have to scale it to at least 50% to see any dirt or pencil remarks on the screen, that is not more than 100dpi (and it's really difficult to clean the correct pixels in a smaller zoom view). So for me it takes the time to scroll thru the whole page, see the dirt and do the cleaning.
I wish I could reach your 5 seconds (even my scanner needs a much longer sweep time) - I would never be able to do any postprocessing in 5 seconds. Please tell me your secret.
Even if the score I scan is good enough that I could omit the manual cleaning I always correct the page size and the position of the score on the page (borders) to the original values. This alone takes me about a minute for each page. Do you think it's wrong do do this?

[daphnis]

I share your view - althou I would say you can really see a difference between 300 and 600dpi when you print the score

That is highly dependent on several factors, including 1.) size of the source material, 2.) size of the paper on which it is to be printed, 3.) resolution of the printer itself and 4.) clarity of the scan/method of printing. Most times material will be printed on paper that is similar to its source, or even slightly smaller, and if you use Acrobat's built-in fit-to-page method the upscaling is more than adequate for 600dpi material (pocket orchestral scores). In fact I only recently attended a rehearsal where I had to print and use an orchestral score from pocket format printed on letter-sized paper, and experienced no trouble.

The reason why I'm still scanning with 1200dpi is, that the filter I use needs at least more than one pixels to work propper in cleaning the scan from dust pixels. Now, when I work the filter on a 600dpi scan I loose to much information of the original score (e.g. some notes tails get lost).

This is why there's really no good replacement for manual post-processing. Most image filters aren't optimized for music, so if you're dealing with a source printed using engraved plates where staccato dots became under inked, many filters will remove this because it doesn't contain enough pixels to be regarded as a normal character. And if you find most of your post-processing is cleaning "salt and pepper" effects (ie dust and dirt, or paper material artifacts), then you should be using a lower threshold. Higher threshold for lower resolution, and vice versa (not considering paper acidity and age qualities). Always remove pencil marks from the source material before scanning as the time it takes to erase is almost always considerably less than manual clean-up.

I wonder how you manage a manual post-processing of the scan in only a few seconds.

I try to minimize the amount of artifacts present on the scanned image by creating a custom profile for each work using random pages as test samples. This allows me to set a proper crop box and threshold levels that cut down on dust/paper acid marks, so my post-processing (in Photoshop with custom droplets and hotkeys assigned) usually only consists of zooming to 25% (correcting for pixel aspect ratio if necessary) and removing any black areas that slipped through (occasional spines, non-exposed areas of the bed where the paper is under-sized, etc.). With the hotkeys and droplets I've assigned, I can usually do the necessary clean-up in 5 seconds or less.

Even if the score I scan is good enough that I could omit the manual cleaning I always correct the page size and the position of the score on the page (borders) to the original values. This alone takes me about a minute for each page. Do you think it's wrong do do this?

This depends. If you want to produce the absolute best quality score for on-screen viewing only, then probably so. For me, as long as the logical page data is set to the right skew, that's all I care about, since music systems can be all different sized, it doesn't tax the eyes to view the next system slightly more left or right than the previous. As for printing, again, if you'll use the Acrobat printing feature "fit to page margins", it will move the image around and center it for you. This is what you want in the printed quality and personally I don't waste time (except on rare occasions) moving the logical page around.

In short, if your scanner supports custom profiles, spend more of your time choosing one that reduces the amount of unwanted pixels before you even go to post-processing. It'll save you heaps of time.

[ilynov]

I would say that 300 dpi is really enough for A4 scores, except for full scores with notes too small. For these, as well as for A5 scores 600 dpi is enough. Bigger resolutions do not, in fact, enhance the quality of the scan - technically.
I use the 4990 Epson scanner with automatic paper feeding. It takes about 15 seconds to scan one page b&w with 300 dpi, about 25 secs with 600 dpi.
When I use manual scanning, it takes me about 20 seconds with 300 dpi.

[Piupianissimo]

I would say that 300 dpi is really enough for A4 scores, except for full scores with notes too small. For these, as well as for A5 scores 600 dpi is enough. Bigger resolutions do not, in fact, enhance the quality of the scan - technically.
I use the 4990 Epson scanner with automatic paper feeding. It takes about 15 seconds to scan one page b&w with 300 dpi, about 25 secs with 600 dpi.
When I use manual scanning, it takes me about 20 seconds with 300 dpi.

Perhaps it depends on the printer used or simply how you want your scores to look like. I really like the scans to be good enough to reproduce the original (and 300 dpi is not sharp as the original, only 600 dpi can make it). You are lucky, that your scanner is better quality than mine - it seems to be at least twice as fast. I think I made some mistakes with the times I wrote on the "High quality scanning" page. It took me sometimes 15-30 minutes to clean my scores from dust when I scanned at 600 dpi. With the 1200 dpi scanning I have the chance to use the filter and this saves me time in manual cleaning - I never need more than 5 minutes now. Of cause I can do without manual cleaning, but there is still some dirt left that I don't like to have on a page. I don't mind that my scanner needs 2 minutes for a 1200 dpi page. It's no job and there is no need for me to produce thousands of pages each day. I'm happy I don't need that much time for manual cleaning any more since I use that filter. I also tried different color profiles and played around with the threshold, but the results don't get much better - perhaps I used already good values before. When the dirt on the page has the same darkness (or even the same color) like the notes there will be no automatism to get rid of it (as daphnis said). One way is cleaning it and the other leave it as it is.
P.S. BTW, you tell me about fast scanning times, I've just found two of your scans on IMSLP ;o)
P.P.S. Good joke daphnis made, 4 seconds to scan a page... I cant even turn the page and put the score onto the scanner surface within that time - am I getting old???

[daphnis]

Good joke daphnis made, 4 seconds to scan a page

This is why I'd rather invest some money in a good scanner. There's nothing more aggravating and counterproductive than fighting with one's own equipment. The newer HPs and Epsons are even faster than my current setup...

[Piupianissimo]

There's nothing more aggravating and counterproductive than fighting with one's own equipment.

But still it's not only the scanner, but you are faster than me.
Yes you're right, I also don't like poor equipment, but the next I would invest in was a A3+ printer to get fine big sized printouts that I can bind in good quality.