http://trove.nla.gov.au/ndp/del/article/59720040
The OCR quality of this article is superb, but the funny thing is that a large chunk of the original is illegible.
I've attached a screen cap (prior to my correcting the text) where you can see that the start of each line is pretty much impossible to read. Fiddling with levels in photoshop I can see that some of that area is virtually blank - apart from some remnant pixels here and there the colour is exactly the same as the background.
There must be further image processing done between the OCR process and web presentation? It's not just compression, it is almost completely blanking out some words.
Check my corrections on the article ( http://trove.nla.gov.au/ndp/del/correction/59720040 ) to see how good the OCR was - many lines perfect, others with a few stray characters, a missing trailing hyphen, or other minor errors. I wouldn't have been able to correct this article otherwise, since I can't read many of the words.
Is Trove able to "reprocess" the scan in any way to make it more legible?
edit: attach doesn't seem to be working, here is the screen capture mentioned above. I've lined up the OCR and original article columns so you can compare them directly:
![]()



Reply With Quote
