PURPOSE: To efficiently and simultaneously execute the segmentation and the recognition of a character by moving a segmenting visual point to an input character pattern, based on an output state of a neural net, whenever the neural net becomes a balanced state.
CONSTITUTION: A neural net 3 of a character segmenting and recognizing device is constituted of a saccade layer 31, a dynamic layer 32, and an output layer 33. The output layer 33 outputs a result of recognition recognized by a neural net. The saccade layer 31 generates a saccade vector for moving a visual point. The dynamic layer 32 takes the balance of the saccade layer 31 and the output layer 33. The saccade layer 31, the dynamic layer 32, and the output layer 33 are constituted by coupling mutually neuron elements for executing a linear sum-of-products operation. A moving means 4 moves a segmenting visual point to an input character pattern 2, based on an output state of the neural net, whenever the neural net becomes a balanced state. In such a way, the segmentation and the recognition of a character can be executed efficiently.