Optical character recognition (OCR)
As a written character is simply a pattern that needs to be recognised, almost every properly programmed pattern matching software can be used for character recognition. Key features to consider are the ability to learn new fonts easily and which basic pattern recognition algorithm is used. Contour based tools work well on text with clear backgrounds and can provide scale and rotation invariance with no reduction in speed. On the other hand contrast based tools provide a more robust detection in poor or changing contrast applications.
Not all OCR tools can cope with proportionally spaced fonts. In the example shown on the right image, the bounding boxes of some of the characters overlap, potentially causing problems.
Special algorithms can be taught that the changing information caused by partial overlap of the bounding boxes is irrelevant for the recognition. This also enables the algorithm to differentiate well between foreground and background, so that complex background patterns such as used in security print have no impact on the OCR.
OCR and polar unwrap
Polar unwrap is a technique that can simplify OCR when the text is not in a straight line. In this first image example the centre and the inside and outside edge of the coin are marked.
The second image shows the border of the coin 'unwrapped' into a straight line, thus enabling the OCR to be performed. This is even possible when the OCR tool does not support scale and rotation invariance.