On Dec 14, 2009, at 12:32 PM, Philip Pemberton wrote: [dave stands up and cheers] -- Dave McGuire Port Charlotte, FL

On Dec 14, 2009, at 12:59 PM, Philip Pemberton wrote: :-) <off-topic rant> I find it incredibly irritating when people use JPEG for things that it's not suited for. It irritates me because it points to the bigger problem of people running full steam ahead and using technologies that they know NOTHING about, and they're just fine with that. Then they try to do things like present schematic diagrams (Yes, schematics! Even supposedly "technical" people do this!) in JPEG, and wonder why they get lots of unsightly artifacts around lines and sharp color transitions. Recently, I heard someone complaining about those artifacts after exporting an AutoCAD drawing to JPEG. He said "oh, they always do that, I think my computer has a virus or something." Moron. I wanted to hit him in the head with a shovel. So, folks: JPEG is for continuous-tone images, like natural scenes and photographs. It DOES NOT WORK even remotely well for "artificial" images. Use PNG or GIF for those. You'll get better compression ratios and MUCH better image quality. This isn't just an anal-retentive academic theoretical difference...we're talking about images that look good versus images that look like crap. </off-topic rant> -Dave -- Dave McGuire Port Charlotte, FL

Lack of proper knowledge.

On Mon, Feb 15, 2010 at 12:14:04AM -0600, Michael B. Brutman wrote: If all you are doing black&white scans, there is an easier way: - scan to bilevel TIFF, ideally compressed using fax G4 - feed the TIFF to tumble[0] to get compact bilevel PDF For my own archival scanning: - scan at 600 dpi to bilevel TIFF - convert TIFF to use fax G4 compression - tar up the TIFF files for archival - feed the TIFF files to tumble to produce PDF for archival - feed the TIFF files to a conversion chain that emits DJVU for archival The scans are pretty compact and I end up with the data in three different open formats, so the odds of being able to read them a few years down the road should be reasonably good ;-) Side benefit: due to the rather high resolution, the prints done from the archival PDFs tend to be much better than Xerox copies (printing with a 600 dpi laserjet). And the resolution is high enough to get reasonable results for OCR, when needed. Kind regards, Alex. [0] http://tumble.brouhaha.com/ -- "Opportunity is missed by most people because it is dressed in overalls and looks like work." -- Thomas A. Edison

Jules Richardson wrote: The problem isn't Imagemagick, it's that when you scan a greyscale image and then downconvert it to 1bpp, you get either black or white. The greyscales disappear and any photos basically get turned to crap. I've been using Xsane to scan -- For greyscale: Res: 600DPI BPP: 8 For B&W: Res: 600DPI BPP: 1 Halftoning: None Brightness: 0 Sharpness: 0 Gamma correction: default Threshold: 128 All saved to TIFF format -- greyscale images get saved as 8bit Deflated TIFF data, B&W as CCITT Group 4 FAX. Then to deal with the moire and screening patterns and convert the greyscale images to PNGs, I feed them through ImageMagick: convert FrontPage-0001.tiff -adaptive-blur 4x4 \ -units PixelsPerInch -density 600 -resample 300 \ -format png FrontPage-0001.png Then finally, convert to PDF with Tumble: tumble -b %F FrontPage-0001.png FrontPage-0002.tiff 0*.tiff \ -o tumbled_doc.pdf And view the result: evince tumbled_doc.pdf I'm not using Imagemagick for the PDF conversion, I'm using Tumble. Also, FYI -- if anyone wants to play with my branch of Tumble (with PNG+JP2 support and the TIFF Photometry Tag bugfix), it's online here: http://hg.philpem.me.uk/tumble/ This syncs against my private Hg repository roughly every five minutes. Either clone the repository (if you use Mercurial you already know how to do this) or go to that URL and click the "bz2" link. @Eric: if you want this fork of Tumble to disappear, let me know off-list. I only forked it because it appeared to be completely unmaintained. Which these are. They're fresh out of the HP shrink-wrap. These seem to be laser-printed. Or if not, they were printed with something very similar to a laser printer -- the pages have that distinctive toner-on-paper feel. Yeah, this is pretty boring. I'm basically scanning a few pages, reading emails, scanning some more pages, doing some coding, and carrying on like that. At least it delays the effects of boredom a little. Cheers, -- Phil. classiccmp at philpem.me.uk http://www.philpem.me.uk/

On Thu, 2009-12-17 17:06:36 +0100, Jan-Benedict Glaw <jbglaw at lug-owl.de> wrote: Hmpf, the script was stripped. Here it is, inline: #!/usr/bin/env sh NETPBM=pbm TESSLANG=deu HOCR2PDF=/home/jbglaw/paper/hocr2/HocrConverter.py # 2 = pixel/inch, 600x600 DPI TIFF_296=2 TIFF_282=600 TIFF_283=600 OUTDIR="__out__.$$" rm -rf "${OUTDIR}" mkdir "${OUTDIR}" # Split mkdir "${OUTDIR}/split" for i in *.tiff; do TIFFNAME="`basename "${i}" | sed -e 's/\.tiff$//'`" tiffsplit "${i}" "${OUTDIR}/split/${TIFFNAME}-" done # Convert to bw/grey/colour NetPBM format for unpaper mkdir "${OUTDIR}/singlepage_pbm" for i in "${OUTDIR}"/split/*.tif; do TIFFNAME="`basename "${i}" | sed -e 's/\.tif$//'`" convert "${i}" "${OUTDIR}/singlepage_pbm/${TIFFNAME}.${NETPBM}" done # Beautify scans mkdir "${OUTDIR}/singlepage_unpaper" for i in "${OUTDIR}"/singlepage_pbm/*."${NETPBM}"; do PNMNAME="`basename "${i}" | sed -e 's/\.'"${NETPBM}"'$//'`" unpaper --pre-rotate 90 --input-pages 1 --output-pages 2 --layout double "${i}" "${OUTDIR}/singlepage_unpaper/${PNMNAME}-%d.${NETPBM}" done # OCR all the stuff mkdir "${OUTDIR}/hocr" for i in "${OUTDIR}"/singlepage_unpaper/*."${NETPBM}"; do PNMNAME="`basename "${i}" | sed -e 's/\.'"${NETPBM}"'$//'`" tesslanguage="${TESSLANG}" export tesslanguage ocroscript rec-tess --tesslanguage="${TESSLANG}" "${i}" > "${OUTDIR}/hocr/${PNMNAME}.hocr" done # Generate TIFFs from beautified NetBPM images mkdir "${OUTDIR}/unpaper_tiffified" for i in "${OUTDIR}"/singlepage_unpaper/*."${NETPBM}"; do PNMNAME="`basename "${i}" | sed -e 's/\.'${NETPBM}'$//'`" convert "${i}" "${OUTDIR}/unpaper_tiffified/${PNMNAME}.tiff" tiffset -s 296 "${TIFF_296}" "${OUTDIR}/unpaper_tiffified/${PNMNAME}.tiff" tiffset -s 282 "${TIFF_282}" "${OUTDIR}/unpaper_tiffified/${PNMNAME}.tiff" tiffset -s 283 "${TIFF_283}" "${OUTDIR}/unpaper_tiffified/${PNMNAME}.tiff" done read -p "Fix images now and press enter" FOO # Generate single page PDF files mkdir "${OUTDIR}/single_pdf" for i in "${OUTDIR}/unpaper_tiffified"/*.tiff; do TIFFNAME="`basename "${i}" | sed -e 's/\.tiff$//'`" "${HOCR2PDF}" "${OUTDIR}/hocr/${TIFFNAME}.hocr" "${i}" "${OUTDIR}/single_pdf/${TIFFNAME}.pdf" done # Generate final PDF pdftk "${OUTDIR}/single_pdf"/*.pdf cat output "${OUTDIR}/book.pdf" MfG, JBG -- Jan-Benedict Glaw jbglaw at lug-owl.de +49-172-7608481 Signature of: Tr?ume nicht von Dein Leben: Lebe Deinen Traum! the second :

On 21 Dec 2009 at 20:35, Jerome H. Fine wrote: I kind of like the idea of using Google to do my work: http://www.labnol.org/software/convert-scanned-pdf-images-to-text- with-google-ocr/5158/ But if you search OCR PDF, you'll get lots of choices, including some free online services. --Chuck

On 21 Dec 2009 at 20:35, Jerome H. Fine wrote: Jerome, As the link Chuck gave you talked about, there's a big difference depending on how the PDF's were created originally. You need to find out if the PDF was created from images, or created from Word or some other text editor(or text source.) You can tell the difference by trying to select some text in Adobe Reader. Right-click the document and pick the "Select Tool." Then try to select some text, if the whole document turns blue (selected), then it was converted from a scan, otherwise you can probably just cut and paste. If converted from a scan, you'll need to OCR it. I've had pretty good results with machine printed code, but given the age/scan quality/print quality can vary the results considerably. See my previous post on what I use to OCR. Keith

On Dec 21, 2009, at 9:38 PM, Al Kossow wrote: I spent a little time playing with ocropus and then teseract, trying to scan pdp-11 diags back to text. I didn't have good luck. I'd be interested if others have a working formula. I did have a little fun "training" tereract on the line printer font. I think that technique holds promise but it needed more data to do a good job (my initial sample was too small, but did improve things a lot). just curious if anyone else has tried training one of the ocr programs to read line printer fonts. -brad

Al Kossow is CORRECT!!!!!!!!!! Look for /pdf/dec/pdp11/tsxPlus/listings/6.40/ at bitsavers. That was a GREAT job Al. THANK YOU! The original text files were lost. ALL of the PDF files are text!