Method of and apparatus for pattern recognition and method of creating pattern recognition dictionary

A method of pattern recognition, comprising the steps of:

• (a) creating N subregions by dividing a region in which a pattern being an object of recognition is pictured into N blocks and numbering the N blocks so that each block has a corresponding subregion number;
• (b) constructing a feature vector having N features as elements by calculating a feature for each of the N subregions in accordance with a prescribed criterion;
• (c) constructing a total of (N-1) feature sets by retrieving the largest or smallest of the N elements of the feature vector to construct a first feature set consisting of one subregion number corresponding to the retrieved element, then retrieving the two largest or two smallest elements to construct a second feature set consisting of a combination of two subregion numbers corresponding to the two retrieved elements, and continuing such a retrieval and construction pattern until constructing the (N-1)th feature set consisting of a combination of (N-1) subregion numbers;
• and
• (d) by reference to a dictionary in which feature sets obtained by performing the steps (a) to (c) on various kinds of model patterns are prerecorded along with category names of the model patterns, obtaining similarities between the feature sets of the input pattern and the feature sets stored in the dictionary in corresponding relationship to the category names, and determining a category name that provides the greatest similarity, as the category name of the input pattern;
• wherein dividing the region in step (a) means dividing the region into N subregions such that the sum of the values of dots is equal for any of the N subregions, and the feature calculated in step (b) is based on the area size of each individual subregion.

An apparatus for pattern recognition, comprising:

• pattern input means for inputting a pattern to be recognized;
• means for creating N subregions by dividing a region in which a pattern being an object of recognition is pictured into N blocks and numbering the N blocks so that each block has a corresponding subregion number;
• means for constructing a feature vector having N features as elements by calculating a feature for each of the N subregions in accordance with a prescribed criterion;
• means for constructing a total of (N-1) feature sets by retrieving the largest or smallest of the N elements of the feature vector to construct a first feature set consisting of one subregion number corresponding to the retrieved element, then retrieving the two largest or two smallest elements to construct a second feature set consisting of a combination of two subregion numbers corresponding to the two retrieved elements, and continuing such a retrieval and construction pattern until constructing the (N-1)th feature set consisting of a combination of (N-1) subregion numbers;
• means for storing a dictionary in which feature sets obtained through processing performed on various kinds of model patterns by the pattern input means, the subregion creating means, the feature vector constructing means, and the feature set constructing means, are prerecorded along with category names of the model patterns;
• and
• means for obtaining similarities between the feature sets of the input pattern and the feature sets stored in the dictionary in corresponding relationship to the category names, and for determining the category name that provides the greatest similarity, as the category name of the input pattern;
• wherein the subregion creating means divides the region in which a pattern being an object of recognition is pictured into N subregions of equal area size, and the feature vector constructing means calculates the feature based on the sum of the values of dots in each individual subregion.

An apparatus for pattern recognition, comprising:

• pattern input means for inputting a pattern to be recognized;
• means for creating N subregions by dividing a region in which a pattern being an object of recognition is pictured into N blocks and numbering the N blocks so that each block has a corresponding subregion number;
• means for constructing a feature vector having N features as elements by calculating a feature for each of the N subregions in accordance with a prescribed criterion;
• means for constructing a total of (N-1) feature sets by retrieving the largest or smallest of the N elements of the feature vector to construct a first feature set consisting of one subregion number corresponding to the retrieved element, then retrieving the two largest or two smallest elements to construct a second feature set consisting of a combination of two subregion numbers corresponding to the two retrieved elements, and continuing such a retrieval and construction pattern until constructing the (N-1)th feature set consisting of a combination of (N-1) subregion numbers;
• means for storing a dictionary in which feature sets obtained through processing performed on various kinds of model patterns by the pattern input means, the subregion creating means, the feature vector constructing means, and the feature set constructing means, are prerecorded along with category names of the model patterns;
• and
• means for obtaining similarities between the feature sets of the input pattern and the feature sets stored in the dictionary in corresponding relationship to the category names, and for determining the category name that provides the greatest similarity, as the category name of the input pattern;
• wherein the subregion creating means divides the region in which a pattern being an object of recognition is pictured into N subregions such that the sum of the values of dots is equal for any of the N subregions, and the feature vector constructing means calculates the feature based on the area size of each individual subregion.

A method of creating a pattern recognition dictionary, comprising the steps of:

• (a) creating N subregions by dividing a region in which a pattern being an object of recognition is pictured into N blocks and numbering the N blocks so that each block has a corresponding subregion number;
• (b) constructing a feature vector having N features as elements by calculating a feature for each of the N subregions in accordance with a prescribed criterion;
• (c) constructing a total of (N-1) feature sets by retrieving the largest or smallest of the N elements of the feature vector to construct a first feature set consisting of one subregion number corresponding to the retrieved element, then retrieving the two largest or two smallest elements to construct a second feature set consisting of a combination of two subregion numbers corresponding to the two retrieved elements, and continuing such a retrieval and construction pattern until constructing the (N-1)th feature set consisting of a combination of (N-1) subregion numbers;
• and
• (d) storing the feature sets in a designated memory by associating each of the feature sets with a category name of the model pattern;
• wherein dividing the region in step (a) means dividing the region into N subregions of equal area size, and the feature calculated in step (b) is based on the sum of the values of dots in each individual subregion.

A method of creating a pattern recognition dictionary, comprising the steps of:

• (a) creating N subregions by dividing a region in which a pattern being an object of recognition is pictured into N blocks and numbering the N blocks so that each block has a corresponding subregion number;
• (b) constructing a feature vector having N features as elements by calculating a feature for each of the N subregions in accordance with a prescribed criterion;
• (c) constructing a total of (N-1) feature sets by retrieving the largest or smallest of the N elements of the feature vector to construct a first feature set consisting of one subregion number corresponding to the retrieved element, then retrieving the two largest or two smallest elements to construct a second feature set consisting of a combination of two subregion numbers corresponding to the two retrieved elements, and continuing such a retrieval and construction pattern until constructing the (N-1)th feature set consisting of a combination of (N-1) subregion numbers;
• and
• (d) storing the feature sets in a designated memory by associating each of the feature sets with a category name of the model pattern;
• wherein dividing the region in step (a) means dividing the region into N subregions such that the sum of the values of dots is equal for any of the N subregions, and the feature calculated in step (b) is based on the area size of each individual subregion.