Fuji Xerox has developed new optical character recognition software that uses algorithms and logic patterns that are quite similar to how the human brain processes the images that it sees.
Reading Japanese is hard, but if you've yet to see actual handwritten Japanese, then you're still oblivious to the language's true nightmarish form. Fortunately, Fuji Xerox holds the potential to make OCR systems a lot better at analyzing characters and letters of such complex languages with their newest developed software.
Fuji Xerox has recently developed new software that uses algorithms and analysis patterns not much different from how your eyes and brain analyzes and recognizes shapes and patterns. The actual scanning and analysis process is not much different to a standard OCR system fundamentally, but it has one crucial added element: adaptation and variable pattern registry. The new software is capable of adapting and learning actively from a multitude of handwritten versions of the same character, just like how our brain registers and records patterns and shapes for future reference.
The process is simply explained by the report in two steps. First, the system extracts all curves, bumps, and bended lines visible on the character, building a composite "map" based on the information extracted. Then, like a standard OCR system, it tries to scan certain recognizable patterns and shapes, until it finally analyzes the entire character. These two steps continuously combine and intertwine in the software's "embed operation" (registry of new patterns) and "sub-sampling" (cross-referencing of multiple similar patters) sections, until a correct analysis is achieved.
The system is claimed to be capable of recognizing almost more than 30,000 characters from different languages, though it is potentially more useful (and practical) for languages that have writing systems of considerable complexity. With its claimed recognition accuracy, it can be well considered as the very first software or system that is capable of performing at such level of precision.