Potential impact of Google Vision API OCR upgrade

Google Vision API OCR is Cloud based, and typically, it is upgraded to a newer version according to Google's release cycle. As a result of this upgrade, an impact might be observed on document extraction processing and extraction results in IQ Bot.

Impact on document classification in IQ Bot after the update of Google Vision API OCR

The classification of documents is based on the OCR result, particularly the order in which the fields are detected. Thus, if the content of the OCR results varies, then the classification results of the similar document processed previously might show visible differences during document extraction.

The auto-upgrade affects the classifier service in two ways:
  • Different groups are assigned to documents: The documents when processed will be assigned to different groups versus what they were originally routed to so far. Consequently, some data extractions might fail, which were previously successful.
  • New groups are created: The documents previously being processed with a group might now result in new group being created.

Impact on data extraction in IQ Bot after update of Google Vision API OCR

A change in OCR output might result in the following differences during data extraction from a similar document type:

  • Improvement in character recognition and its associated impact: As a result of improved keyword character recognition, previously unrecognized values from a document will now provide better extraction results.
    Note: If any processing logic is written to fix the lack of detected characters, it might cause issues when the document is processed.
  • Degradation in character recognition and its impact: Validation might fail due to degradation in OCR output, and the document will be routed to the Validator. In a scenario where there is no validation setup, the extracted data output might see a degradation.

Options to mitigate

If you encounter any of the previously mentioned issues, you have the following options to explore:

  • Retrain existing learning instance to fix issues with document classification.
  • Write processing logic using Python to mitigate extraction outcomes.
  • Evaluate creating learning instance with different OCR provider supported by IQ Bot.
  • Raise a support ticket with Automation Anywhere.