摘要 |
PROBLEM TO BE SOLVED: To precisely recognize a pattern by taking and collating also information on the size and shape of an individual pattern. SOLUTION: Many model patterns are previously sorted into respective classes and a scale and shape expressing the feature of each class are selected class by class to prepare a model pattern vector. A feature extraction matrix is obtained from the prepared model pattern vector and from the feature extraction matrix, its representative feature amount vector is obtained to be stored in a learning means 4. Next, a feature amount vector is extracted by using the feature extraction matrix from the means 4, the similarity of this and the representative feature amount vector from the means 4 is calculated, an input pattern is estimated from this similarity and information on the size and shape of the pattern are effectively taken out to extract features.
|