Intelligent Recognition Technology

The Final Frontier in Text Recognition

The basic principle of Parascript® Intelligent Recognition states that handwriting, when reduced to its most basic components, is essentially motion, or a series of movements, made by a writing instrument. According to this theory, any handwriting can be described using elements of a special description language. The eight elements that make up the trajectories of all cursive letters (Figure 1 below) form a ring that illustrates the possible transitions of neighbor elements. 

Figure 2 - An example of the letter “d” described using motion theory. The order of elements in the letter description follows the trajectory of a pen. Horizontal lines show the vertical position on the image associated with each element in the letter description.

Principles of Dynamic Intelligent Recognition

Both OCR and ICR deliver high accuracy when analyzing constrained text (OCR with machine print and ICR with handprint) but are ineffective when dealing with cursive, where letters are linked together, and may be poorly written or even illegible. Consider a situation where the symbol segmentation of an image is ambiguous. In Figure 3 below, an OCR/ICR recognition system could determine that the first symbol is a “d”or a combination of a “c” and an “l”. Depending on the segmentation, the reading result produced by a letter-based recognition technology may be completely different: “clear” in the first case and “dear” in the second.  

As accurate character segmentation is critical, Intelligent Recognition can often recognize poor-quality text that would be impossible for OCR and ICR systems to recognize. Intelligent Recognition dynamically uses context – in a process similar to the one humans employ when reading and interpreting text – to compensate for the inherent ambiguity of human handwriting. The context is used during the recognition process rather than after recognition, when results might already have been misinterpreted, thus improving the accuracy of results. Again, going back to Figure 3, it is not clear if the first symbol is a “d” or a combination of a “c” and an “l”.

The dynamic vocabularies contained in Intelligent Recognition systems do not analyze and store all possible hypotheses of segmentation. If the dynamic vocabulary does not contain a combination of “c” and an “l” at the beginning of the word, the only possible segmentation solution is “d”. The dynamic usage of context eliminates all impossible combinations from the solution set, enabling the evaluation of results “on the fly” during the recognition process. Dynamic context, therefore, provides the highest possible recognition accuracy, because it eliminates the impossible results in real time, during the recognition process.

The Final Frontier in Text Recognition

Intelligent Recognition technology often recognizes text that is considered to be of poor quality or even completely unacceptable for OCR and ICR technologies, therefore further improving the recognition rates when compared to other systems.

Working with high quality machine print, OCR provides recognition accuracy of nearly 100 percent (99.9 %), a level of accuracy acceptable for many forms processing applications. ICR cannot guarantee the same levels of accuracy that OCR systems deliver on machine print due to the inherent problems of reading handprint – spacing variations, diversity of human writing styles, etc. Instead, state of the art ICR systems provide the same recognition accuracy for a certain part of the data stream, while the data that cannot be reliably read continue to be sent for visual verification. The following mechanism is used by ICRs to ensure the accuracy required by the application. The stream of images is divided into two parts: those that were recognized reliably with a required accuracy (accepted), and those for which the system does not guarantee the required accuracy (rejected).

Intelligent Recognition further improves recognition rates and accuracy when compared to traditional machine print (OCR) and handprint (ICR) engines through field recognition and cross-validation of results.

Field Recognition

Intelligent Recognition recognizes a field not a character, and consequently a whole field is either accepted or rejected. Conversely, in the case of a rejected field Intelligent Recognition technology additionally provides information about unreliable characters. Second, the reject mechanism is tuned so thoroughly that it allows accuracy up to 0.1% for the texts of low quality.

Cross-validation of Results

Computing power alone is not able to deliver high recognition results without a human-like recognition approach. Intelligent Recognition employs the most advanced methods of single character recognition while using sophisticated algorithms to cross-validate results during the recognition process.

Intelligent Recognition advances the state of recognition technology, exploiting the strengths and capabilities of its predecessors – OCR and ICR systems – while eliminating their inherent limitations. Intelligent Recognition technology delivers highly accurate machine print, handprint and cursive recognition results, helps eliminate laborious human data entry and has become a proven solution for a broad range of the most demanding applications for government posts, commercial mailers, banks and financial institutions, BPO and data processing centers.