PURPOSE: To convert digital synapse input signals timely matched with each other into an analog form for being impressed to a neural network layer.
CONSTITUTION: Several digital synapse input signals are written in a digital memory at different time, and timely matched and read in order to be converted into an analog form for being impressed to a neural network layer. The digital synapse input signals are outputted from a device 2 which operates the raster scanning of the field of a video sample or the successive field, and the neural network layer 3 has a filter for recognizing the shape in a visual field of the video sample. The neural network layer 3 is a preliminarily programmed form or learning form, and the action output is digitized for a digital processing.