3.12 OCR

Vendor should use auto-rotate and voting when generating OCR. Most OCR software offers an auto-rotate option. When auto-rotate is enabled, the software will OCR each image four times, rotated 90 degrees each time. It determines the best result and publishes the content to the load file. The majority of documents have the same orientation: portrait. Without auto-rotate, these documents can yield good results. The rest of the documents may be designed for a landscape layout, such as an HR chart. Other documents still may have been scanned “upside-down”, resulting in garbage OCR. OCR voting is a process where multiple OCR programs compare results to determine the best results.

Quality Check

The OCR text should best approximate and recreate the formatting found on the original image. The OCR field should never be just the words in one long string.

No text and the top, bottom or either side should be clipped.

Multi-Page Text Files

There should be a one document to one OCR text file ratio. The OCR filename must match the document image key. So, a 10 page document with the image key of AA001 should have a corresponding file AA001.TXT that contains the OCR for AA001 through AA010.

Each page of OCR should have a line identifying the page number, or Bates number. In this fashion, people can search for any Bates number and find the correct document. Please include space between the OCR text and page marker.

The following shows sample OCR:

<< AA001 >>

Text for first page

<< AA002 >>

Text for second page

The following chart shows a sample database and corresponding OCR files:

* Please refer to Bates prefix and suffix conventions.