Neocognitron as a tool for optical marks recognition
Oleksii Gorokhovatskyi · 2016
Paper describes an investigation of simplified neocognitron neural network model as a tool for practical recognition of handwritten mark images. Simplification of neocognitron structure from only two stages and fixed number of feature-extraction planes is proposed, the overall stages of solving practical image processing problem are described. Recognition properties of simplified net are investigated, it is shown that for particular task simple structure of net may be the same effective (up to 98% of correct recognition) as traditional four-stages one.