OCR against photos

[stephanhodges]

For fun, I loaded a screen shot of a dictionary entry. I have my system set for Simplified.

Is it true that the OCR won't do both simplified and traditional at the same time?

If so, is there a quick on-screen button/setting to toggle between the two?

OCR handled "yi4yi4" (simplified), but didn't get the traditional characters in square brackets.

Attached both source photo and OCR results.
SOURCE:


RESULT:

 

[scykei]

It works in iOS though.

 

[mikelove]

Is it true that the OCR won't do both simplified and traditional at the same time?

No, works fine for both.

OCR handled "yi4yi4" (simplified), but didn't get the traditional characters in square brackets.

Combination of the thickness of the Android system font and the close spacing of its characters, I think - in general if we were actually designing an OCR system to recognize screenshots we'd calibrate it differently.

 

[scykei]

Don't you think that OCR in a screenshot seems like something most people would want to do? I know I've done it a few times and I do admittedly struggle with it. I have to put the box around the word I want to translate. Ideally, I would like to have the full page OCR thing on and just tap on anything I don't know.

I mean it works, but if you said you could optimise it better... Maybe incorporate PDF OCR detection system elsewhere too?

 

[mikelove]

Don't you think that OCR in a screenshot seems like something most people would want to do? I know I've done it a few times and I do admittedly struggle with it. I have to put the box around the word I want to translate. Ideally, I would like to have the full page OCR thing on and just tap on anything I don't know.

Screenshots and PDFs / scanned documents are very different things - OCR should work quite well with PDFs (it does in our testing anyway), as the fonts / character spacing / stroke widths / etc of those are designed around print rather than screen viewing, but actual screenshots are a different story since they tend to use very different fonts / spacing / etc. Honestly, I have yet to hear any compelling use case for OCRing screenshots - if the text already exists in electronic format then OCR seems like a decidedly inefficient way to go about translating it. Are you trying to get text out of some other app that won't let you copy-and-paste?

 

[stephanhodges]

For me, normally, I would take photos of things and then run OCR against those. Since I have a 10" tablet, I thought the photos and the screen shots were the same resolution. Obviously wouldn't be the case for smaller devices, however.

I don't know what would be involved in optimization for this type of thing, but OCR IS one of the strong selling points of Pleco, outside of core dictionaries, etc.

 

[mikelove]

For me, normally, I would take photos of things and then run OCR against those. Since I have a 10" tablet, I thought the photos and the screen shots were the same resolution. Obviously wouldn't be the case for smaller devices, however.

Oh, no, photos are much larger on most devices. Though it's really more an issue of fonts and stroke weights than of resolution.

I don't know what would be involved in optimization for this type of thing, but OCR IS one of the strong selling points of Pleco, outside of core dictionaries, etc.

Sure, but it's a much more risky thing to bet the business on, which is why I've generally tried to avoid doing so - OCR will likely get much more ubiquitous and cheap as augmented reality becomes more mainstream.

 

[scykei]

There are a lot of instances where you cannot copy the text, like Appstore descriptions for Chinese apps, menu labels, etc. Also, we occasionally see people who post screenshots of text online for whatever reason.

No need to spend a lot of time optimising it though. It works very well as is, since I've been using it a few times for this purpose.

 

[Alexis]

Isn't the issue here the type of font being OCR'd, not whether the source is a photo or a screenshot? I have succesfully OCR'd Screenshots of epub/pdfs with no problem.

 

[mikelove]

Isn't the issue here the type of font being OCR'd, not whether the source is a photo or a screenshot? I have succesfully OCR'd Screenshots of epub/pdfs with no problem.

Mostly, yes - in this case we had a font with punctuation that was similar in size / stroke width, which you don't generally see in printed Chinese text.