Announcing a Dead-Simple Command-Line Tool for Text Recognition

9 points

6 years ago

I find it unnecessarily difficult to do text recognition on images and documents.

Either it's too cumbersome to set up or too hard to work with the output. Or worse, the text recognition itself fails. (I'm looking at you, tesseract.)

The only service I've found useful is Google Cloud Vision. I do admit the on-boarding process is somewhat cumbersome, but considering how phenomenal the accuracy is and how usable the output is, Google Cloud Vision is worth it.

I want to make text recognition 1) less cumbersome to set up and 2) even easier to work with the output, all while maintaining the phenomenal accuracy of Google's solution.

The culmination of my work is a command-line tool which, I believe, achieves those goals. Simply download the executable and specify the files you want to recognize text in. You get the results as JSON:

  $ ./sight receipt_1.pdf invoice.png -o results.json --prompt-api-key

The tool [1] is open source and ships alongside a Go client for the API, which I expose for developers. The API simply expects POST requests of base64-encoded files and returns recognized text as JSON.

Since this is a service, my startup [2] does charge for it: $0.50 per 1,000 pages --- one third the price of Google Cloud Vision and Amazon Textract (both charge $1.50 per 1,000 pages). But every new account gets 1,000 pages for free, to convince you of the quality, speed, and ease of use!

1. https://github.com/siftrics/sight

2. https://siftrics.com/