Edit

Switch ABBYY OCR API from FRE11 to FRE12

note

In order to improve OCR recognition for specific use cases, an additional OCR Engine FRE12 is supported starting from Intelligent Automation Cloud Enterprise 10.1.

important

Starting from Intelligent Automation Cloud Enterprise 10.1, both versions of SDK - FRE Engine 11 and FR Engine 12 - are supported.

ABBYY FineReader Engine SDK upgrade to version 12 resulted in a number of improvements, enabling our users to select between FRE SDK 11 and FRE SDK 12. Thus, we added a second (alternative) OCR engine, with advanced recognition of tables and texts in images, as well as significantly improved recognition of Japanese characters.

Switching from FR Engine 11 to FR Engine 12 and vice versa is available in the following ways.

API changes

For using FRE12 via API, you need to change the existing v1 prefix to v2.

• v1 - FRE11 functionality is still actual and available for use
• v2 - FRE12 functionality is actual and available for use

FRE11 example of API URLs:

https://<ocr_hostname>/api/v1/cloud/processImage

https://<ocr_hostname>/api/v1/cloud/activateLicense

FRE12 example of API URLs:

https://<ocr_hostname>/api/v2/cloud/processImage

https://<ocr_hostname>/api/v2/cloud/activateLicense

BP changes

The OCR plugin uses FRE11 by default. Using FR Engine 12 (API v2) in BP since Intelligent Automation Cloud Enterprise 10.1 will require OCR API v2 URL to be specified:

• as a new variable in Data Store > Global Variables
• also specified individually for each corresponding Bot Task by using a proper OCR API v2 URL defined in Global Variables

To configure a Global Variable to use with FR Engine 12 (API v2), do as follows.

1. Add a new variable ocr_url_v2 and define URL for OCR API server used with FR Engine 12.

2. Keep original ocr_url in the Global Variables data store to use FRE11 (API v1) as a default OCR engine for existing Bot Tasks when upgrading or making a new installation in a "classic" way.

   

3. Make sure Bot Tasks designed to use with new FR Engine 12 have ocr_url_v2 instead of ocr_url. To define ocr_url_v2, go to your Business Process > Edit Workflow > select a Bot Task > Design > the Create Bot Configuration section.

   

It is possible to combine FRE11 (API v1) and FRE12 (API v2) in BP for some use cases. To do this, a Bot Task requires to have default ocr_url OR ocr_url_v2 specified in a Bot Configuration section accordingly to the OCR engine that needs to be used.

important

Mind that each Bot Task needs to have only one OCR URL API specified, whether it should be v1 or v2.

OCR URL in step results

To display ocr_url in step results, you can use the export plugin at the end of the Bot Task.

<export include-original-data="true">
<single-column name="ocr_url" value="${ocr_url}"/>
</export>

FAQ

What OCR engine and OCR API will be used in Intelligent Automation Cloud Enterprise 10.1 by default?

FR Engine 11 (API v1)—the same stable and well-known OCR engine we have used since version 8.x.

Why can't we just switch to FRE12 in Intelligent Automation Cloud Enterprise 10.1 if it's said to be better than FRE11?

(1) At the current moment FRE11 and FRE12 are more complementary than interchangeable. FRE11 is a reliable OCR engine that works fine in many cases. Thus, we won't recommend FRE12 as a replacement for the current OCR engine until we collect enough feedback from PS, COE, and our partners. Consider it as a beta available since 10.1.

(2) FRE12 produces results incompatible with previously trained models used with FRE11.

At the moment, we can't provide an effortless way to retrain models to be used with FRE12. In general, FRE12 performs better (improvements vary 5-20% for different tests) and could be the only solution when it comes to CJK docs (especially modern Japanese), documents with complicated layouts, tables, texts over stamps, etc. It is still highly recommended to perform tests and choose which of FRE-s works best for a specific BP.

How long FRE11 will be supported by Intelligent Automation Cloud Enterprise and other products?

FR Engine 11 is at the end of its product lifecycle according to its vendor (ABBYY). Only some of the critical issues are fixed. FRE11 will be deprecated for the next major release, so it will be only available for backward compatibility for the previously trained models. We expect other teams to advocate for FRE12 for each new client and each new trained ML model.