28 Nov
2025
28 Nov
'25
8:57 a.m.
Greetings Restorers,
I think a number of us have wanted to restore software that's only
available as a scanned listing from a line printer. The original
printout probably wasn't the best typographic quality, and scanning
doesn't improve it.
As a first pass, OCR with tools like Adobe Acrobat can easily produce
a rough draft of the content in text form, but it takes almost as much
work to correct the many "typos" as it does to simply re-type the listing.
It seems like, with all this high-tech AI processing around, it
should be possible to take advantage of the limited character set, fixed
fonts, and restricted grammar that one might find in a listing to
resolve more of the ambiguities in character recognition.
Does anyone have an approach that's more efficient than generic OCR
and a long process of correcting typos on every line of code or comment?
Thanks
/guy
28 Nov
28 Nov
9:32 a.m.
I use a simple script that uploads each file to chatgpt and collects the
transcript. It works shockingly well compared to all ocr solutions.
Dne pet., 28. nov. 2025 ob 15:57 je Guy Fedorkow via cctalk <
cctalk(a)classiccmp.org> napisal/-a:
Greetings Restorers,
I think a number of us have wanted to restore software that's only
available as a scanned listing from a line printer. The original
printout probably wasn't the best typographic quality, and scanning
doesn't improve it.
As a first pass, OCR with tools like Adobe Acrobat can easily produce
a rough draft of the content in text form, but it takes almost as much
work to correct the many "typos" as it does to simply re-type the listing.
It seems like, with all this high-tech AI processing around, it
should be possible to take advantage of the limited character set, fixed
fonts, and restricted grammar that one might find in a listing to
resolve more of the ambiguities in character recognition.
Does anyone have an approach that's more efficient than generic OCR
and a long process of correcting typos on every line of code or comment?
Thanks
/guy
30 Nov
30 Nov
3:42 p.m.
I've been trying to OCR the lineprinter listing of the 9815 calculator. I
tried a few OCRs and the most accurate was onlineocr.net . It's still full
of errors and limited to 15 pages per day, but may be useful for some needs.
On Sat, Nov 29, 2025 at 6:51 AM bostjan spetic via cctalk <
cctalk(a)classiccmp.org> wrote:
I use a simple script that uploads each file to
chatgpt and collects the
transcript. It works shockingly well compared to all ocr solutions.
3:53 p.m.
If you have the paper copy, you can help the ocr’ing by using a pen and tracing lightly
missing characters.
It helps quite a bit and is fast to do.
Sent from my iPhone
On Nov 30, 2025, at 13:42, Adrian Godwin via cctalk
<cctalk(a)classiccmp.org> wrote:
I've been trying to OCR the lineprinter listing of the 9815 calculator. I
tried a few OCRs and the most accurate was onlineocr.net . It's still full
of errors and limited to 15 pages per day, but may be useful for some needs.
> On Sat, Nov 29, 2025 at 6:51 AM bostjan spetic via cctalk <
> cctalk(a)classiccmp.org> wrote:
>
> I use a simple script that uploads each file to chatgpt and collects the
> transcript. It works shockingly well compared to all ocr solutions.
>
>
4:19 p.m.
No, it's a scanned listing from a patent. I think it's on the Australian ho
museum site. But that does make me wonder if preprocessing the input image
files would be faster than postprocessing the text output files.
On Sun, 30 Nov 2025, 21:53 Wayne S, <wayne.sudol(a)hotmail.com> wrote:
If you have the paper copy, you can help the ocr’ing
by using a pen and
tracing lightly missing characters.
It helps quite a bit and is fast to do.
Sent from my iPhone
On Nov 30, 2025, at 13:42, Adrian Godwin via
cctalk <
cctalk(a)classiccmp.org> wrote:
I've been trying to OCR the lineprinter listing of the 9815 calculator.
I
tried a few OCRs and the most accurate was
onlineocr.net . It's still
full
of errors and limited to 15 pages per day, but
may be useful for some
needs.
> On Sat, Nov 29, 2025 at 6:51 AM bostjan spetic via cctalk <
> cctalk(a)classiccmp.org> wrote:
>
> I use a simple script that uploads each file to chatgpt and collects the
> transcript. It works shockingly well compared to all ocr solutions.
>
>
4:25 p.m.
He don’t like they say, garbage in, garbage out.
Pre-processing will make things much faster and cleaner.
You might want to acquire one of those cheap HP desktop scanners with a feed and then
print out the patent code on 8 1/2 x 11 paper and go over it.
Then rescan it and see if it makes a difference.
Sent from my iPhone
On Nov 30, 2025, at 14:19, Adrian Godwin <artgodwin(a)gmail.com> wrote:
No, it's a scanned listing from a patent. I think it's on the Australian ho museum
site. But that does make me wonder if preprocessing the input image files would be faster
than postprocessing the text output files.
On Sun, 30 Nov 2025, 21:53 Wayne S,
<wayne.sudol@hotmail.com<mailto:wayne.sudol@hotmail.com>> wrote:
If you have the paper copy, you can help the ocr’ing by using a pen and tracing lightly
missing characters.
It helps quite a bit and is fast to do.
Sent from my iPhone
On Nov 30, 2025, at 13:42, Adrian Godwin via cctalk
<cctalk@classiccmp.org<mailto:cctalk@classiccmp.org>> wrote:
I've been trying to OCR the lineprinter listing of the 9815 calculator. I
tried a few OCRs and the most accurate was onlineocr.net<http://onlineocr.net> .
It's still full
of errors and limited to 15 pages per day, but may be useful for some needs.
> On Sat, Nov 29, 2025 at 6:51 AM bostjan spetic via cctalk <
> cctalk@classiccmp.org<mailto:cctalk@classiccmp.org>> wrote:
>
> I use a simple script that uploads each file to chatgpt and collects the
> transcript. It works shockingly well compared to all ocr solutions.
>
>
4:29 p.m.
Another thing to do, is if you print it out and you have a copier at work, is enlarge the
printout a bit to make it more legible, then go over it with pen and rescan it.
Sent from my iPhone
On Nov 30, 2025, at 14:25, Wayne S <wayne.sudol(a)hotmail.com> wrote:
He don’t like they say, garbage in, garbage out.
Pre-processing will make things much faster and cleaner.
You might want to acquire one of those cheap HP desktop scanners with a feed and then
print out the patent code on 8 1/2 x 11 paper and go over it.
Then rescan it and see if it makes a difference.
Sent from my iPhone
On Nov 30, 2025, at 14:19, Adrian Godwin <artgodwin(a)gmail.com> wrote:
No, it's a scanned listing from a patent. I think it's on the Australian ho museum
site. But that does make me wonder if preprocessing the input image files would be faster
than postprocessing the text output files.
On Sun, 30 Nov 2025, 21:53 Wayne S,
<wayne.sudol@hotmail.com<mailto:wayne.sudol@hotmail.com>> wrote:
If you have the paper copy, you can help the ocr’ing by using a pen and tracing lightly
missing characters.
It helps quite a bit and is fast to do.
Sent from my iPhone
On Nov 30, 2025, at 13:42, Adrian Godwin via cctalk
<cctalk@classiccmp.org<mailto:cctalk@classiccmp.org>> wrote:
I've been trying to OCR the lineprinter listing of the 9815 calculator. I
tried a few OCRs and the most accurate was onlineocr.net<http://onlineocr.net> .
It's still full
of errors and limited to 15 pages per day, but may be useful for some needs.
> On Sat, Nov 29, 2025 at 6:51 AM bostjan spetic via cctalk <
> cctalk@classiccmp.org<mailto:cctalk@classiccmp.org>> wrote:
>
> I use a simple script that uploads each file to chatgpt and collects the
> transcript. It works shockingly well compared to all ocr solutions.
>
>
3 Dec
3 Dec
8:42 a.m.
On Nov 30, 2025, at 14:19, Adrian Godwin
<artgodwin(a)gmail.com> wrote:
No, it's a scanned listing from a patent. I think it's on the Australian ho
museum site. But that does make me wonder if preprocessing the input image files would be
faster than postprocessing the text output files.
If the scan you have is not very good, an option would be to contact the patent office for
a copy. It may well be a copy direct from their archives is cleaner than what you have
now.
paul
9:55 a.m.
I don't think it's the general quality of the patent print that's poor,
it's the line-printer listing section from
https://www.hp9845.net/9845/downloads/patents/US4089059.pdf starting at
about page 213 of the pdf , possibly section 26 of the patent.
The print in that section is much paler than the rest - typical of a worn
line-printer ribbon. I doubt the printed copy is any better. I'm only
trying to OCR the listing, not the rest of the patent.
On Wed, Dec 3, 2025 at 2:42 PM Paul Koning via cctalk <cctalk(a)classiccmp.org>
wrote:
On Nov 30, 2025, at 14:19, Adrian Godwin
<artgodwin(a)gmail.com> wrote:
No, it's a scanned listing from a patent. I think it's on the Australian
ho museum site. But that does make me wonder if preprocessing the input
image files would be faster than postprocessing the text output files.
If the scan you have is not very good, an option would be to contact the
patent office for a copy. It may well be a copy direct from their archives
is cleaner than what you have now.
paul
1:55 p.m.
On Dec 3, 2025, at 10:55 AM, Adrian Godwin via cctalk
<cctalk(a)classiccmp.org> wrote:
I don't think it's the general quality of the patent print that's poor,
it's the line-printer listing section from
https://www.hp9845.net/9845/downloads/patents/US4089059.pdf starting at
about page 213 of the pdf , possibly section 26 of the patent.
The print in that section is much paler than the rest - typical of a worn
line-printer ribbon. I doubt the printed copy is any better. I'm only
trying to OCR the listing, not the rest of the patent.
That's quite a cleaen listing, actually, cleaner than most I have worked with and
dramatically better than some. The sort of slightly-damaged characters that appear should
be no problem at all for the "training" feature of ABBYY Fine Reader to deal
with. What you'd have to do is run a number of pages through it in training mode, so
it sees a number of variations of the individual characters. And as I mentioned,
you'd do all the scanning in the mode where it only accepts what it was trained with,
no "builtin" patterns. That way it won't make up stuff that isn't part
of the character set but happens to match something built-in, like a pound-sterling sign.
It may be that scanning the listing as a table (with the various columns as table columns)
will work well, and give you the layout explicitly. Or it can be scanned as plain text,
but in that case the spacing will mostly turn into individual spaces and you'd need
post-processing to insert tabs etc. to make it look right again. Given the simple
assembler syntax involved that sort of post-processing would not be hard.
paul
1 Dec
1 Dec
4:11 p.m.
Nov 30, 2025, 13:42 by cctalk(a)classiccmp.org:
I've been trying to OCR the lineprinter listing of
the 9815 calculator. I
tried a few OCRs and the most accurate was onlineocr.net . It's still full
of errors and limited to 15 pages per day, but may be useful for some needs.
If it helps, I have an online OCR application that you're welcome to try:
https://keston.drwho.virtadpt.net/
It's not perfect but for whatever it's worth I keep it updated with a daily
cronjob to catch
changes to the image models.
--
The Doctor [412/724/301/703/415/510]
WWW: https://drwho.virtadpt.net/
"Nine Tesla, and the ghosts were everywhere."
4:51 p.m.
Thanks, the project is on the backburner a bit due to other pressures but
it would be great to try both that and the preprocessing sometime and try
to move it along a bit. The idea is to OCR the listing and cross-check it
with a rom dump (this will also add the maths routines which were not
included in the listing).
On Mon, Dec 1, 2025 at 10:20 PM The Doctor via cctalk <cctalk(a)classiccmp.org>
wrote:
Nov 30, 2025, 13:42 by cctalk(a)classiccmp.org:
If it helps, I have an online OCR application that you're welcome to try:
https://keston.drwho.virtadpt.net/
It's not perfect but for whatever it's worth I keep it updated with a
daily cronjob to catch
changes to the image models.
--
The Doctor [412/724/301/703/415/510]
WWW: https://drwho.virtadpt.net/
"Nine Tesla, and the ghosts were everywhere."
I've been trying to OCR the lineprinter
listing of the 9815 calculator. I
tried a few OCRs and the most accurate was onlineocr.net . It's still
full
of errors and limited to 15 pages per day, but
may be useful for some
needs.
5:19 p.m.
Adrian, a great line-by-line comparison tool is Beyond Compare from “Scooter Software”
It’s free.
Sent from my iPhone
On Dec 1, 2025, at 14:52, Adrian Godwin via cctalk
<cctalk(a)classiccmp.org> wrote:
Thanks, the project is on the backburner a bit due to other pressures but
it would be great to try both that and the preprocessing sometime and try
to move it along a bit. The idea is to OCR the listing and cross-check it
with a rom dump (this will also add the maths routines which were not
included in the listing).
> On Mon, Dec 1, 2025 at 10:20 PM The Doctor via cctalk <cctalk(a)classiccmp.org>
> wrote:
>
>
> Nov 30, 2025, 13:42 by cctalk(a)classiccmp.org:
>
>> I've been trying to OCR the lineprinter listing of the 9815 calculator. I
>> tried a few OCRs and the most accurate was onlineocr.net . It's still
> full
>> of errors and limited to 15 pages per day, but may be useful for some
> needs.
>>
> If it helps, I have an online OCR application that you're welcome to try:
>
> https://keston.drwho.virtadpt.net/
>
> It's not perfect but for whatever it's worth I keep it updated with a
> daily cronjob to catch
> changes to the image models.
> --
> The Doctor [412/724/301/703/415/510]
> WWW: https://drwho.virtadpt.net/
> "Nine Tesla, and the ghosts were everywhere."
>
>
>
28 Nov
28 Nov
9:56 a.m.
I have a copy of Abbey Fine Reader Pro which I got free on a magazine
many years ago.
If it reads a character incorrectly you can add to the image <=>
character map so it can adapt for example to a damaged slug on a line
printer train or other type element.
Its not 100% but I used it to scan the IBM1130 CSMP from the manual....
Dave
On 28/11/2025 14:57, Guy Fedorkow via cctalk wrote:
Greetings Restorers,
I think a number of us have wanted to restore software that's only
available as a scanned listing from a line printer. The original
printout probably wasn't the best typographic quality, and scanning
doesn't improve it.
As a first pass, OCR with tools like Adobe Acrobat can easily
produce a rough draft of the content in text form, but it takes almost
as much work to correct the many "typos" as it does to simply re-type
the listing.
It seems like, with all this high-tech AI processing around, it
should be possible to take advantage of the limited character set,
fixed fonts, and restricted grammar that one might find in a listing
to resolve more of the ambiguities in character recognition.
Does anyone have an approach that's more efficient than generic OCR
and a long process of correcting typos on every line of code or comment?
Thanks
/guy
12:35 p.m.
I too have used Fine Reader (paid for, but the price is not all that high) to OCR various
listings. It's way better than the Acrobat OCR, at least the one I tried ages ago on
the Ethernet spec (back before Bitsavers, or at least before I found it). The learning
feature is very good, and important when dealing with low quality input. You can also
tell it only to recognize what you taught it, i.e., not try to match any builtin patterns.
If you're doing line printer listings with 64 character sets, that's helpful,
otherwise it may mistake something blurry for a pound-sterling sign.
In some cases OCR just can't hack what it is given and the only option is to type it
all in again. I've done that with some listings that were ugly enough I sometimes had
to zoom in to have one letter take up most of the screen, just to figure out what it was.
In that particular case, it also used a lot of overprinting to deal with mixed case text:
upper case letters represented lower case, upper case with dot overprint for actual upper
case. You'd thing that OCR training could handle that, but the difference is subtle
enough it doesn't really work.
Supposedly current versions of the open source program Tesseract are pretty good, but I
haven't tried it. Looking for how to train it got me all confused, it didn't seem
to be something that was at all convenient, not like the interactive training feature of
Fine Reader.
OCR likely will not handle fixed layout well (not unless you can treat it as tables). If
that's important, some Python or Emacs post-processing can clean up a lot. Similarly
if there are common recognition errors you can spot by pattern matching. Scanning the CDC
6600 wire lists goes well this way, because the data have a very consistent pattern. For
example, the OCR might mix up zero and oh, but an edit pass can fix those 100%.
paul
On Nov 28, 2025, at 10:56 AM, David Wade via cctalk
<cctalk(a)classiccmp.org> wrote:
I have a copy of Abbey Fine Reader Pro which I got free on a magazine many years ago.
If it reads a character incorrectly you can add to the image <=> character map so
it can adapt for example to a damaged slug on a line printer train or other type element.
Its not 100% but I used it to scan the IBM1130 CSMP from the manual....
Dave
On 28/11/2025 14:57, Guy Fedorkow via cctalk wrote:
Greetings Restorers,
I think a number of us have wanted to restore software that's only available as a
scanned listing from a line printer. The original printout probably wasn't the best
typographic quality, and scanning doesn't improve it.
As a first pass, OCR with tools like Adobe Acrobat can easily produce a rough draft of
the content in text form, but it takes almost as much work to correct the many
"typos" as it does to simply re-type the listing.
It seems like, with all this high-tech AI processing around, it should be possible to
take advantage of the limited character set, fixed fonts, and restricted grammar that one
might find in a listing to resolve more of the ambiguities in character recognition.
Does anyone have an approach that's more efficient than generic OCR and a long
process of correcting typos on every line of code or comment?
Thanks
/guy
3:03 p.m.
On Fri, 28 Nov 2025, Guy Fedorkow via cctalk wrote:
Does anyone have an approach that's more
efficient than generic OCR
and a long process of correcting typos on every line of code or comment?
My wife often gets had written notes or just as worse, photos of
screens[1] from which she has to re-type in the contents.
She uses "Claude" or "ChatGPT" with remarkable efficiency. However
note
that this is written text rather than computer code, so the results may
vary.
Gordon
[1] No. Don't ask why the senders can't send the original document. It's
just not worth it - they're as thick as mince.
1
days inactive
6
days old
15 comments
8 participants
participants (8)
-
Adrian Godwin -
bostjan spetic -
David Wade -
Gordon Henderson -
Guy Fedorkow -
Paul Koning -
The Doctor -
Wayne S