One outcome of NEC Corp’s artificial intelligence work is a character-recognition system using neural network technology, according to Microbytes. The system will be linked to a scanner to create a low-cost character input system for the NEC PC-9801 personal computer series: it passes scanned images to the neural network in the form of 9-by-11-dot character matrices and the network reproduces them, using an error-correction error correction feedback algorithm to improve accuracy. NEC claims that the system recognised a 62-character alphanumeric string from a matrix printer with was 99.8% accuracy and a 76 character typewritten alpha-numeric and symbol string with 99.87% accuracy where the conventional characteristic extraction algorithm managed only 98.5% and 96.67% accuracy on the same tests. Speed, at 50 characters a minute, is about the same with either system. NEC is working to improve the neural network technology, believing it could lead to a system able to read hand-written characters.
