![]() |
OCR on jolla
As not (yet?) on open repos thought would post here:
https://github.com/skvark/Textractor...ses/tag/v0.1-1 Author is looking for icon and banner for harbour (@unpixels on twitter) |
Re: OCR on jolla
Thanks for posting this here, I got the icon and banner already. Next release will be in the Jolla store when I got some things sorted out.
|
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
This really is great (except it throws an error when using screenshots, might need to change output format if possible), having only ancient greek/maths and english it still recognized half of a greek airport sign (english parts in higher percentage), the brussels sign to throw paper INTO the toilet (seemed funny as most of the time it was reverse, don't clog our precious pipes), sadly 0 accuracy, but it was in weird font anyway. When will realtime OCR/translation come? :D
edit: just an extract I find amazing: Code:
Litterin‘g in the terminal is not permitted. Please dispose of cigarette butts,gUm,y 1 https://pbs.twimg.com/media/CGm9PYGWcAIMfo_.jpg (don't ask me why it is turned around, gallery upload to twitter? (photofunia had similar bug with initial pre-4.29 release) or the textractor is impacted, not sure) |
Re: OCR on jolla
Quote:
I'll check why images gets mirrored. |
Re: OCR on jolla
Thanks! (weird thing is that image still displays properly in gallery, but after textractor went through it and gallery upload to twitter it ended up turned sideways (just checked with fresh image from gallery, behaves as expected), yeah some things in jolla work in mysterious ways C: )
|
Re: OCR on jolla
Quote:
I tried to reproduce that mirroring issue by uploading some images taken in Textractor/selected via Textractor to Twitter and everything worked correctly. Did you scale the image before upload in the gallery app and did you take the image in Textractor or did you select the image? |
Re: OCR on jolla
Quote:
edit: yeah, tried to reproduce, but bukowski's quote is oriented correctly (opening the image by the app 'modifies it' as in shuffles to the top of available options, even if nothing was done with it) |
Re: OCR on jolla
Aaaah and got it, picture from mexico I took (no text, just for test:
https://twitter.com/szopin/status/60...479424/photo/1 This got turned left-right, this is the resulting B/W OCR image (already turned) https://twitter.com/szopin/status/60...356288/photo/1 ) |
Re: OCR on jolla
Quote:
Yes, it shuffles to the top of options because it is actually modified even if you see message "An error occured. Image could not be preprocessed." because before preprocessing the image is opened, the dots per meter value is changed and rotation is applied if necessary -> original image is overwritten by these changes. |
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Quote:
edit: yea, old pictures not taken by jolla camera - display/upload correct, taken by jolla camera - display correct in gallery view, turned left-rigth when uploaded to twitter, seems like jolla bug (unless touching by the app is also needed edit: yeah, untouched pictures from gallery display correctly after uploading, so the 'touching' by textractor is involved) |
Re: OCR on jolla
Quote:
Edit: Ah, I made changes so the image is copied before preprocessing - this should fix the problem at least partly. |
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Quote:
this should be untouched by any post processing (outside of jollas) this is DL link for unmodified in any way (I hope) file: https://www.dropbox.com/s/f1p7d8ff3f...5_002.jpg?dl=0 |
Re: OCR on jolla
New release available with some bugfixes: https://github.com/skvark/Textractor...ses/tag/v0.2-1
|
Re: OCR on jolla
Quote:
If someone sees that bug still in some situation, please report. And those who had some photos affected by this, you can mirror/rotate the image back in some image manipulation program. I'm not sure does the Qt's QImage retain exif data after read -> save operation with same name. If not, I'm sorry for any inconvenience caused by this unintended "feature". |
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Quote:
|
Re: OCR on jolla
Textractor is now available in the Jolla Store :cool:
Please report bugs and other stuff (feature requests) to the issues of the Textractor Github repository: https://github.com/skvark/Textractor/issues |
Re: OCR on jolla
Wow. I tried this on a shop receipt and it worked well (language: Finnish). I'm impressed.
|
Re: OCR on jolla
Seems like photos have to be taken in "landscape view". Is that supposed to be like this?
Did some tests and all the pictures I took in portrait mode result in complete rubbish shown after processing... But in correct view, recognition is actually quite good (Language: German). |
Re: OCR on jolla
I've been waiting for an OCR on a mobile device for years! My first exposure to it was when PC/AT with 80286 CPUs running at 16 MHz were all the rage and I found it strange that it had all but disappeared from the surface of the earth.
Now finally my waiting is over! Thank you! Of course, I played with it a bit and must say the results are pretty good. There are a few cases when a glare etc can cause some distortion and spoil the result but that is kinda expected. My favourite was trying to render a shot taken from the computer screen. This is what I got from the shot covering the second and third post in this thread: Code:
If I may have a feature request, I would like to see rendering of some common programming languages. Particularly C and C++. Most of them look like English so it should not be a big problem. |
Re: OCR on jolla
You also maybe should add a .nomedia-file to your app folder because the pictures will show up e.g. in Android Whats**p. :)
|
Re: OCR on jolla
Quote:
They also show in Gallery. Unless Textractor deletes them immediately or provides a means to do that within the app, I prefer to leave it as it is so I can delete them in Gallery. |
Re: OCR on jolla
Why should the app save the images to some other location than in the gallery? As pichlo noted, I'm not going to invent the wheel again if some other app already provides means to for example delete the image.
Quote:
And yes, taking an image of a screen is not going to work so well in most cases, the background is too hard for the preprocessing step unless there are clearly pure black/white regions. |
Re: OCR on jolla
Quote:
Quote:
|
Re: OCR on jolla
Quote:
I will add an feature which allows to lock the orientation in camera mode. |
Re: OCR on jolla
Best results so far when Textractor used combined with Sailfish Scanner to make pictures.
|
Re: OCR on jolla
Quote:
I'm not sure if resuing Sailfish Scanner code is good solution for this since it would introduce additional dependencies to Python and to some Python modules. I believe I can replicate the same functionality with less code utilizing native QTransform class. However, depending on license (I didn't see any), the qml side could be useful to look at. |
Re: OCR on jolla
Excellent. One thing that makes sailfish-scanner an adequate image maker for Textractor (besides its edit features) is that it privileges short exposure time so it reduces the risk of fuzziness. Perhaps a good idea would be to allow short exposure time with ISO value increased accordingly.
|
Re: OCR on jolla
I finally found some time to implement some new features. The work is still ongoing, but currently following new features are implemented:
I'll probably add support for text extraction from PDF files too. Some small enchancements for example to camera behaviour will be done too. |
Re: OCR on jolla
There's now a pre-release available for testing: https://github.com/skvark/Textractor...ses/tag/v0.5-0
If there are no big bugs, I will push next release to the Jolla Store in near future. Changelog: New features:
Bug fixes:
Miscellaneous:
|
Re: OCR on jolla
I added i486 build to GitHub: https://github.com/skvark/Textractor...ses/tag/v0.6-0
Can someone who has the tablet test it? I can't test the app completely on emulator. This version is identical to the phone version, no changes in code. Locked in portrait mode and doesn't look very good on the tablet. I'll add this version to the store if it works correctly. I will enhance the tablet UI when I have time. |
Re: OCR on jolla
Quote:
Any specific test you would want me to try? One usability proposal: When cropping those corner points -> they are really close to edge so its difficult to move them without app starting to minimize. So those "balls" could be little bit "in" all ready to make cropping easier. |
Re: OCR on jolla
Quote:
Yes, I have noticed the same usability issue and I'll fix it :) I just noticed also that Harbour validation fails due to different allowed requires on i486 target (vs. armvhl7) so I have to make some changes to the app anyway before I can submit it to the store. |
Re: OCR on jolla
2 Attachment(s)
Hello,
I take a picture from a Poker machine on CRT screen. I want to extract all Counters from this machine to text format. I select French and English languages but No/bad results. Check attached pictures. Why it's not possible to have all numbers correctly? Thanks for help. |
All times are GMT. The time now is 08:07. |
vBulletin® Version 3.8.8