Show HN: An API to extract texts from images and PDF files

we think that's easier to integrate as there is nothing to install, nothing to maintain. We are also doing some pre/post processing which allow orientation detection for example. We also think that's might be useful for mobile as they are limited in resources. Also, on a PDF, we have an hybric way to perform both to not use slow OCR methods on easy to extract text but that's still quite a young product. New advanced features are coming.

thx, I am gonna fix that

The upgrade button doesn't work, and nobody is going to hover long enough to see the "Not Available Yet" title. And the current 10 requests isn't even enough to test with.

I'm excited to try this.. so figure out a way to take my money soon.

I come bearing gifts, if anyone would like to host some of this themselves.

This follows the API documented by Stampin (minus the throttling errors)--it does not currently do the OCR, but as mentioned elsewhere by zdw you can probably get tesseract to get you like 80% of the way there. If you wanted to use that, you'd likely just replace the hacky `pdftotext` callout with your preferred toolchain.

You'll need Ruby, Sinatra, and the Xpdf tools, I believe.

Dual-licensed under the AGPL, BSD, and WTFPL licenses. idklol.

The code:

  require 'sinatra'
  require 'json'

  use Rack::Logger

  post '/extracttext' do

      begin
      status 204 and return unless params["file"] != nil

      type = params["type"] || "text"
      lang = params["lang"] || "en"

      tmpfilename = params["file"][:tempfile].path
      `pdftotext #{tmpfilename}`
      File.delete(tmpfilename)

      convfile = File.open("#{tmpfilename}.txt","r")
      lines = convfile.read.split("\n")
      convfile.close
      File.delete(convfile.path)

      content_type "application/json"
      {"text"=>lines}.to_json

      rescue
          status 500 and return
      end
  end

EDIT:

For God's sake run this in a jail and only on an internal network!

The OCR is really useless. I tested it with some reddit "advice animal" memes (because there is a need for transcriptions). You would think that text is pretty simple and easy, but the output I got was like:

    /\n\nnmrs wn\ufb02qyi mm mm\nTlIIEI\ufb02|\ufb02llllM\u2018l co