How Does ICR Work?

The technology behind ICR is just an extension of OCR technology. Both types of software function by trying to match the characters presented to them to existing characters from a predefined database. However, because ICR works with text that is less precise, the process is more complex and less perfect. While both OCR and ICR rely on statistical methods, ICR also relies on a semantic approach that uses contours to identify the characteristics of a text character. For example, the software can identify a part of a character as an arc, a loop, a line up, or a line diagonally down, among other contour descriptions. If each character is broken down sufficiently, there will eventually be enough information to match it to a character that is known to the program.

To perform ICR on a document, the process is the same as for OCR. You must first place a document on a scanner, so that the image is captured, usually as a TIFF file. Then, the software aligns the image properly to match ICR zones with expected data fields, so that it can identify zones that need to be processed. Sometimes, documents and forms may contain zones with handwritten text as well as zones with typed text. These need to be separated and processed using different tools.

Next, the software may get rid of lines and boxes that surround ICR zones, and despeckle the image to make it clearer. Once the image is ready, the ICR recognition engine assesses the characters and assigns confidence values to each character. Depending on the program�s level of confidence, additional validation procedures may be required, including spell-checking and matching information to preexisting data tables. Even after this has been done, it is possible that errors remain. Therefore, the software identifies characters that are still uncertain, and these need to be confirmed or corrected manually. This last step is a very important part of the ICR process, because it reduces the occurrence of errors that are instantly noticeable for a person.