Classification dictionary learning system, classification dictionary learning method and recording medium

A classification dictionary learning method used for a classification dictionary learning system for generating a classification dictionary, the classification dictionary learning method comprising:

• receiving a priority degree of a classification between a first class and a second class, the second class including one or more sub-classes;
• based on the value that corresponds to the priority degree, setting a weight value of a weight table in which a weight coefficient of each of a plurality of learning samples is described for each of the sub-classes;
• the priority degree;
• learning a plurality of classifiers on the basis of the priority degree for each combination of the sub-classes;
• selecting a combination of the sub-classes in which a specific optimization index is the most preferable with regard to each of the classifiers;
• and
• updating a weight on the basis of a classification accuracy based on the classifier learned with the selected combination for setting the priority degree of the classification between the first class and the second class to a classification among the subclasses included in the second class.

A non-transitory computer-readable recording medium storing a classification dictionary learning program used for a classification dictionary learning system for generating a classification dictionary, wherein the classification dictionary learning program causes a computer to perform functions that include:

• receiving a priority degree of a classification between a first class and a second class, the second class including one or more sub-classes;
• based on the value that corresponds to the priority degree, setting a weight value of a weight table in which a weight coefficient of each of a plurality of learning samples is described for each of the sub-classes;
• learning a plurality of classifiers on the basis of the priority degree for each combination of the sub-class;
• selecting a combination of the sub-classes in which a specific optimization index is the most preferable with regard to each of the classifiers;
• and
• updating a weight on the basis of a classification accuracy based on a classifier learned with the selected combination for setting the priority degree of the classification between the first class and the second class to a classification among the subclasses included in the second class.