Reverse Video & OCR

Reverse video detection is important for high fidelity OCR recognition. Many color documents have text that is lighter (rather than darker) than its background. There are many OCR engines that rely on certain assumptions, like text being darker than background. This is a limitation in several commercial OCR products that are fairly highly rated, yet do not perform well given a reverse video recognition test.

In the reverse video recognition test, a set of color image documents where the text is consistently darker than the background is given to an OCR engine. With this "video" set of document scans a baseline recognition rate is established for the OCR system. Next, the reverse color mapping (aka color complement) of the original images is computed. This 2nd image set is now considered and the engine's OCR accuracy for these reverse video documents is determined. If the OCR's system performance on the reverse video documents is statistically worse than on the original set then this OCR engine is not invariant to reverse video transformations and is biased towards relatively well-behaved dark text on light background. Obviously, an OCR recognition system that has no such bias (e.g., CVISION's OCR) is more highly rated.

Click here to read next topic: OCR & Novel Fonts

Return to Table of Content