Optical character recognition web app in JS and HTML5 | Hacker News Reader

Optical character recognition web app in JS and HTML5

kdzwinel.github.io

28 pointskonradzikusek19 comments

I tried it with 10 writings from documents, books, printed ads, papers and so on. Even crystal clear text with only one character yielded only /"////,,,// gibberish. I found that it doesn't work very well...

That's strange, I never got that bad results. Check out my video where I'm demonstrating the app ( https://www.youtube.com/watch?v=ttn437BlEbo ), you'll see what texts I've worked on.

Is there functionality to train the algorithm? Improve the answer by using Levenshtein distance between OCRed text and set of words or something? Or maybe that is something the dev who uses this library can implement themselves. :)

Using simple text correction algorithm should work just fine. However, this goes beyond this simple experiment. I just wanted to demonstrate capabilities of ocrad.js . Feel free to implement it yourself! It seems that there are multiple JS auto-correction libs, e.g. https://code.google.com/p/careti/ .

Could use a bit of tweaking. Maybe my camera image isn't good enough?

http://www.tiikoni.com/tis/view/?id=e6fe518

I took a picture of text on the side of a mug and got it to work. Just had to play with the brightness and contrast. Pretty cool, thanks for sharing.

Nice, but you should make it work automatically and "online" just from camera image, without setting additional parameters.

That would be great! Any idea how to locate the text on the image? Since face recognition can be done "online" ( http://auduno.github.io/clmtrackr/examples/facesubstitution.... ), finding text should also be possible.

Works dandy nice job! Why does it have to be a 'huge' image? Can't the cropped area be zoomed before running the OCR?

Thanks! It doesn't have to be 'huge' image, I just recommended it as it gives better results. Small text tends to be more distorted.

Doesn't work very well for me in terms of recognition. I took a picture of a few logos and big text and it only recognized 80%

Recognition algorithm is provided by ocrad.js which is a JavaScript version of the OCRAD project ( http://www.gnu.org/software/ocrad/ ). It's not perfect, but IMO it's very good - just make sure to use huge, printed, black-on-white text.

Could this be made to work on the iPhone ?

It should probably work on Chrome for iPhone. It won't work on Safari since it doesn't support getUserMedia and WebGL.

You could use trigger.io or another phone frame work that gives you JS api access to the camera and pipe it to this library

Got it to recognize 3 out of 6 letters on my beer bottle. Not bad!

Only minified JS? Open source? So disappointing.

I'm not sure what do you mean. Only code that I wrote for this demo is in index.html and main.js. These two files are not minified. Ocrad.js, jquery, modernizr, bootstrap are minified because these are libraries that I'm importing. Everything is explained under the app, please read the description.

Sorry, I misunderstood it. :( I browsed too fast.

Reply on news.ycombinator.com