PATTERN IDENTIFICATION METHOD AND PATTERN IDENTIFICATION DEVICE

摘要

PROBLEM TO BE SOLVED: To provide a pattern identification method and a pattern identification device capable of achieving a reduction of a calculation cost by downsizing a network in a pattern identification using a hierarchical neural network having a characteristic detection neuron without lowering accuracy of processing.

SOLUTION: A partial characteristic detection part D23 obtains an input unit of a neural network based on the characteristic of an image detected by an initial processing part D22. A property determination processing part D26 extracts a candidate of a pattern to be identified from the center of the image. A number of couplings reduction processing part D27 reduces the number of couplings of input units and output units by merging predetermined input units in one. A facial image identification part D28 identifies whether there is a pattern in the image using the input units with the reduced number of couplings.

COPYRIGHT: (C)2005,JPO&NCIPI