Getting better OCR results

  • 3.2K Views
  • Last Post 24 January 2012
  • Topic Is Solved
Vasily Panferov posted this 23 January 2012

I'd like to improve recognition results of the web service. Can you give me some tips?

Vasily Panferov posted this 24 January 2012

General settings

Check the recognition language

First, check the recognition language. If no language parameter is specified, the recognition will be performed using English dictionary and character set. If your documents are in other language or contain a mixture of languages, you need to specify them explicitly using language parameter. There is a possibility to specify more than one language in comma-separated list

processImage?language=Italian,English

processImage?language=Greek&exportFormat=txt

The full list of supported languages is here

Handwritten and handprinted text

Although OCR is able to recognize handwritten text, the best results are obtained if you know the text region in advance and use field-level API with processTextField.

Camera shots

Better lighting conditions

The higher quality you original image is, the better recognition results you'll get. If your images are obtained with a phone camera, try to shot in good lightning conditions, make sure no flash is visible on a shot.

Narrow the search region for the text

If you have a camera that contains little text and much of surrounding environment, the results are unpredictible. If you are taking the pictures, make sure that text occupies large part of resulting image.

  • Liked by
  • Andrey Isaev
Close