System and method for preparing a recognition dictionary

• US 4,658,429 A
• Filed: 12/31/1984
• Issued: 04/14/1987
• Est. Priority Date: 12/29/1983
• Status: Expired due to Fees

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Kobayashi, Yoshiki, Kubo, Yutaka, Orita, Miyahiko
• Primary Examiner(s): Kreitman, Stephen A.
• Application Number:
  US06/687,757
• Time in Patent Office: 834 Days
• Field of Search: 382/36, 382/37, 382/38, 382/6
• US Class Current: 382/225
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers

Abstract

For the preparation of a tree structure recognition dictionary, a feature set at each node of the tree should be the feature giving the largest discrete distribution number. When elements of an object are classified or divided into categories, if the number of combinations of subgroups, which are discrete, is the largest, the classification is the most effective. This classification is effected, when the distance between distributions for a certain feature is larger than a predetermined value.

Claims

1. A recognition system for preparing a recognition dictionary comprising:

means for producing distribution data of a plurality of categories for each of a plurality of features from samples of patterns to be recognized;

means responsive to said means for producing distribution data for providing the mean value and the standard deviation of each of the categories of each of said plurality of features;

means responsive to said means for providing the mean value and the standard deviation for summing for each of the features a number of combinations of pairs of the categories having a discreteness greater than a predetermined value; and

means coupled to said means for summing for comparing the number of combinations of each feature for selecting as a highest priority one of the features of the categories to be recognized having the highest number of combinations.

2. A recognition system according to claim 1, wherein said means for summing sums combinations between two adjacent and discrete categories for each of said plurality of features.

3. A recognition system according to claim 1, further comprising means for providing an interdistribution distance, wherein when the mean value of a first category is larger than the mean value of the second category, said interdistribution distance is defined by the ratio of the difference between the lower limit of said second category as a function of dispersion of the category and the upper limlt of said first category as a function of dispersion of the category to the difference between either one of which is larger between the upper limit of said second category as a function of dispersion of the category and the upper limit of said first category as a function of dispersion of the category and either one which is smaller between the lower limit of said first category as a function of dispersion of the category and the lower limit of said second category as a function of dispersion of the category.

4. A recognition system in accordance with claim 3 wherein when the mean value of a first category is larger than the mean value of the second category, said interdistribution distance is defined by the ratio of the difference between the lower limit of said second category as a function of dispersion of the category and the upper limit of said first category as a function of dispersion of the category to the difference between either one of which is larger between the upper limit of said second category as a function of dispersion of the category and the upper limit of said first category as a function of dispersion of the class and either one which is smaller between the lower limit of said first category as a function of dispersion of the category and the lower limit of said second category as a function of dispersion of the category.

5. A recognition system in accordance with claim 3 wherein said second predetermined value is equal to said first predetermined value.

6. A recognition system in accordance with claim 3 wherein the pattern to be recognized is a set of characters including alphanumerical symbols and each of the features is information involving lightness and darkness of a contour of said characters.

7. A recognition system in accordance with claim 3 wherein the pattern to be recognized are objects having different shapes and each of the features is information involving lightness and darkness of a contour of the object.

8. A recognition system in accordance with claim 1 further comprising:

means comparing a value representing a distance in distribution between a pair of categories of said feature having the highest priority with a second predetermined value.

9. A recognition method for preparing a recognition dictionary comprising the steps:

producing distribution data of a plurality of categories for each of a plurality of features from samples of patterns to be recognized;

providing the mean value and standard deviation of each of the categories of each of the plurality of features in response to the distribution data;

summing for each of the features a number of combinations of pairs of the categories having a discreteness greater than a predetermined value in response to the mean value and standard deviation; and

comparing the number of combinations of each feature for selecting as a highest priority one of the features of the categories to be recognized having the highest number of combinations in response to the summed number of combinations.