OCR Recognition Modes

FineReader Engine
9.x, 10, 11
Technology & Features
Recognition, OCR: Speed & Quality

Accurate, or Normal Mode

  • If the developer does not specify anything, the “normal” recognition mode is used.
  • This mode is the most accurate one, because it uses all of the available character recognition technologies available in ABBYY technology.
  • The speed of the recognition depends on the image quality.
  • Only if you process really bad quality images, it is recommended to turn on the “Low Resolution Mode”

Fast OCR Mode

  • This mode provides 2-2.5 times faster recognition speed at the cost of a moderately increased error rate (1.5-2 times more errors).
  • In the case of a hand-printed text (text type TT_Handprinted), a special recognition mode is used.
  • On good print quality texts, recognition makes an average of 1-2 errors per page, and such moderate increase in error rate can be easily tolerated in many cases, such as full text indexing with “fuzzy” searches, preliminary recognition, etc.
  • FineReader Engine 10 image prepossessing and binarisation got massive improvements. Die to the better image quality that is delivered by the enhanced image pre-processing the fast mode could also be re-adjsuted and made dramatically faster.

Results are based on ABBYY internal tests, they will vary with different document sets, different OCR language and the hardware that was used for processing

Note: ABBYY does not recommend using the fast mode to recognize small image fragments (for example, fragments which consist of only one line or word) because the time advantage will be insignificant.

Balanced OCR Mode

  • The balanced mode is an intermediate mode between full and fast mode recognition.
  • The fast mode can be activated with the help of the FastMode property.
  • This property is available for machine-printed texts only, for hand-printed texts the recognition will be run in full mode.

Low Resolution Mode

  • New special processing mode was introduced in FineReader Engine 10.
  • Additional classifier were trained on low resolution scans and faxes
  • About 20% more accurate for
    • low resolution scans
    • About 5% - for other documents
    • But about 10% slower than Normal mode

Back to Technology & Features

This website uses cookies which enable you to see pages or use other functions of our websites. You can turn off such cookies in your browser’s settings. If you continue to use these pages, you consent to the use of cookies.