Still OCR problem with scrolling screenshots

davylue

秀才
Hello everyone,
After using the still OCR feature for the first couple of time the past few days to read image file blog posts, I decided I could use the scroll screenshot feature of my phone in order to use the OCR feature more efficiently, not having to do countless individual screenshot (the image file is quite long, it's a short story).
Problem: the scroll shot displays with high compression, making the OCR fail at most of the characters. However, when I have a look at the scrolling screenshot in my phone gallery, the quality is perfect, so this has to have sth to do with Pleco. If I use normal sized screenshots the quality is just fine.
I didn't find anything in the settings that could solve my problem. Can somebody help me?
 

mikelove

皇帝
Staff member
We currently cap OCR images at 4000x4000 pixels for performance and RAM reasons, so if it's larger than that it gets scaled down. We could theoretically fix this with some engineering work, but we haven't really seen a reason to do so thus far, since 4000x4000 is plenty big enough for photos of texts (even quite large ones, posters e.g.) and screenshot-stitching isn't something we've had a lot of requests for.

Is there a reason why you can't open these blog posts in Pleco as regular text? Either with our Screen Reader function or by 'share'ing the web page itself with Pleco Reader?
 

davylue

秀才
Thank you for your reply!

The blog posts I am referring to are from Weibo and are in the form of image files. Some of those are just very long pictures, for which I think using the scrolling screenshot feature is the quickest way to capture them. Being image files, I can't just copy the text. Also, I downloaded apps with Chinese (amateur?) novels and another one with mahua, for all of which the copying feature will not work. This is why I think I have to resort to screenshots. Pleco reader didn't appear in the share options for me.

Ideally, it should be possible to enable or disable image scaling in the settings. But I guess it won't kill me to do a couple of extra screenshots hehe.

Thanks.
 

mikelove

皇帝
Staff member
What about our “Screen OCR” feature, in that case? Would at least let you remain in context / make it easy to navigate while you’re looking up words.
 

n9

秀才
Would be nice if you could fix the Screen OCR not working on black text over green background, btw (like in your own WeChat messages). I also had a problem a few times where shutting off the smartphone/letting the smartphone let itself shut off while having the Screen OCR open results in the phone not reacting to anything anymore (won't turn on, was with a Wiko FEVER 5G).
 

mikelove

皇帝
Staff member
That should be working already, we can try to investigate why it's not. Does it work better if you confine the recognition area to just the green-box half of the screen in WeChat?
 

n9

秀才
No, and the characters it recognizes vary wildly when you just slightly adjust the rectangle. Giant characters mixed with smaller ones, with no rhyme or reason. Sometimes including a white background text block with the green one helps.
 

mikelove

皇帝
Staff member
Hmm, yeah, sounds like it's having a hard time binarizing the picture correctly (finding light/dark areas). Does it help if you turn on 'advanced binarization' in the still OCR options?
 
Top