| 摘要 |
<p>A pattern recognition method comprises the steps of: (a) dividing a recognition object region of an input pattern into N region, allocating divition region numbers to the N regions to form N division regions; (b) calculation a feature quantity of each of the N division regions in accordance with a predetermined criteria and generating a feature vector comprising the N feature quantities as the elements; (c) picking up largest or smallest element among the N elements of the feature vector, making up a first feature set comprising one division region number corresponding to the largest or smallest element picked up, then picking up two elements which are the largest and second largest elements or the smallest and second smallest elements, making up a second feature set comprising the two division region numbers corresponding to the two elements, and making up an (N-1)th feature set comprising the (N-1) division regions numbers picked up in the same way so as to make up (N-1) feature sets in total; and (d) referring to a dictionary in which feature sets obtained by executing in advance the steps (a) to (c) for various mode patterns, and the category names of the model patterns are recorded, obtaining the similarities of the feature sets of the input pattern with the feature sets in the dictionary for every category name, and determining the category name for which the similarity is the greatest as the category name of the input pattern. The time taken for recognition can be shortened and the recognition rate can be improved. <IMAGE></p> |
