No description
| JP2 | ||
| JP2-glossary | ||
| JP2-misc | ||
| JP2-notes | ||
| PNG | ||
| PNG-glossary | ||
| PNG-misc | ||
| PNG-notes | ||
| .gitignore | ||
| CC-1.0-Universal.txt | ||
| README.md | ||
Sir Gawain and the Green Knight
Page images.
Source:
Convert
Converting the JP2 files to PNG for OCR.
Grayscale
Tests:
# 0, no gamma
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -define png:compression-level=9 -define png:compression-filter=5 PNG/sirgawaingreenkn00unse_0039.png
# 1, ok but still dark. Total OCR fail though, line numbers everywhere.
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -gamma 0.7 1/sirgawaingreenkn00unse_0039.png
# 2, good contrast. Ok, missed u/v, o/u.
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -contrast-stretch 0 2/sirgawaingreenkn00unse_0039.png
# 3, looks like 2. Better OCR than 2, good u/v, but o/u error. Winner so far.
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -level 5%,95% 3/sirgawaingreenkn00unse_0039.png
# 4, lightest without being blown out. Bad, chars that aren't there, wrong.
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -gamma 0.7 -contrast-stretch 0x1% 4/sirgawaingreenkn00unse_0039.png
# 5, bad, too blown out. OCR really bad.
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -threshold 30% 5/sirgawaingreenkn00unse_0039.png
# 6, 3 with compression: same, but missed one letter worse than 3. Really bad one time.
convert -define jp2:rate=0 JP2/sirgawaingreenkn00unse_0039.jp2 -colorspace Gray -level 5%,95% -define png:compression-level=9 -define png:compression-filter=5 -define png:compression-strategy=1 tmp/6/sirgawaingreenkn00unse_0039.png
Final
Ran this:
for i in sirgawaingreenkn00unse_*.jp2
do echo $i
convert -define jp2:rate=0 $i -colorspace Gray -level 5%,95% ../PNG/$(basename $i .jp2).png
done