Monochrome version of BSB PDFs

[WilliamBunting]

http://imslp.org/wiki/2_Fantasiest%C3%B ... f,_Joachim)
has a poor quality monochrome version made with GIMP and Scan Tailor.

Is there any way to get better results?

[cypressdome]

Further feedback will have to wait until the file clears copyright review but until then perhaps you can tell us what images you used as the source: the BSB PDF, the 100% jpgs (which appear to be what Caprotti's submission is based upon), or the 200% jpgs (edit-just noticed there is also 150% jpgs). You would think the BSB PDF would use the highest quality images but the image dimensions from the PDF are only two-thirds of those of the 200% jpgs and both have the same 72dpi resolution. I haven't processed many BSB scores for submission because of the frustration of dealing with their images. The 200% jpgs (and not all scores on BSB have those available) are around 2000 by 2500 pixels for the Raff work. I've dealt with many archive.org scores of that size but those have a resolution of 300dpi and when they are in focus they can be re-sampled to 600dpi and converted to black and white with great success. At 72dpi the best you can seem to do is re-sample it to 150dpi and have an image that is 4000 by 5000 pixels at 150dpi. Then, convert it to black and white using an application like Gimp, Photoshop, or Imagemagick that allows you to set the threshold level. In my opinion the result is less than ideal because of the hindrance of having to start with such a low resolution image. Would it really make a difference to re-sample a 72dpi image all the way to 600dpi? I don't know, but it seems to me that you are starting out with an image that already has too much data missing for it to be worthwhile. I'm sure others have had better luck with BSB images so let's see what they say.

Cypressdome

[WilliamBunting]

The source was the BSB PDF, the one you get from clicking on "download PDF" at the top right.

[cypressdome]

Images extracted from the BSB PDF will be 1000 by about 1270 pixels at 72dpi. Color images at that resolution will look alright (but not great) on the computer screen but I don't think they'll ever look good converted to black and white. You would be better off starting with the best images BSB offers: the 200% jpgs. The hidden url for those images would start at http://daten.digitale-sammlungen.de/~db ... _00001.jpg with the file name going to up to _00015.jpg. My suggestion would be to re-sample these to 150dpi in Gimp (while the image is still in color), use the threshold tool to convert them to black and white, and save them as tifs. Then, if desired, run them through ScanTailor to de-skew and set uniform margins. You might want to test setting the output dpi in ScanTailor to 150 (effectively no re-sample), 300 (2 times), and 600 (4 times) to see if there is any improvement above the initial Gimp re-sample to 150dpi.

Have fun!
Cypressdome

[kalliwoda]

Do not get confused by the strange dimensions of the BSB jpg's: At 75 dpi the page dimensions are hugely inflated, the 2500x2000 pixel scans correspond to 200 dpi at the original page dimensions. These are not great, but can still be used.

Your file conversion seems to have been done by direct conversion to b/w at an identical resolution, followed by some dust filter. This results in the loss of many grey and thin lines.

Somewhere on imslp I remember simple instructions for conversion, but cannot find them now. Here are the essentials:
1) Color scans contain (obviously) more information than b/w monochrome for the same resolution. This extra information can be used if the color scan is converted to a monochrome of 2-3 times resolution. In practice, this is done by first increasing resolution of the color scan from 200 dpi to 600 dpi, and then converting to monochrome.
2) Conversion to monochrome is rarely of acceptable quality, if you use the 50% or the dither option on the mode menu. But photo-editing software also has an adjustable b/w threshold among the "settings": With a setting of "200" (instead of default "120") these Raff scans will be very readable.
3) Finally cleaning with a filter (at 2 points for the 600 dpi file) will also smooth some of the artifacts of the original 200 dpi scan.
4) Now you can convert to b/w bitmap mode using the 50% setting.
The entire process can also be automated for multiple pages, which have then to be assembled into a multipage pdf.
I will try and upload a better version of the Raff in a few day as demonstration.

I use an older Photoshop version, so I cannot comment on how to do these steps with Gimp or other freeware.
As far as I am aware, multipage color pdf's have to be split into individual pages, before they can be processed, and then reassembled.

[WilliamBunting]

Thank you Kalliwoda for the excellent Raff violin pieces. I'll try to follow your method in future.

The same problem exists with ThULB- see my unusable Ernst Elegie arr Ritter.

A couple of other points: A slightly different problem is found on Rode Air varie arr Ritter from the Barato colllection- it needs despeckling to make it useful.

Finally, I recently acquired Adobe Photoshop Lightroom 4. Will this be useful for music score cleaning?

Regards WB