Recognition device, recognition method, and computer program product

A recognition method comprising:

• performing, plural times, a selection process of selecting a plurality of groups each including one or more learning samples from a learning sample storage unit that stores therein a plurality of learning samples that are classified into a plurality of categories;
• learning a classification metric, which is a classifier Fi(x) for each group of the plurality of groups for classifying the plurality of groups selected in each selection process, where 1≦i≦N, and N is the number of classifiers;
• performing a generation process of generating an evaluation metric that includes the classifier Fi(x) for each group of the plurality of groups, wherein the evaluation metric is defined as {F1(x), F2(x), through, FN(x)}, and where x is a variable;
• receiving an image including an object as a recognition target;
• extracting a first feature value representing an image feature of the image;
• transforming the first feature value into a second feature value using the evaluation metric;
• calculating, by referring to a recognition table stored in a recognition table storage unit in which a plurality of reference feature values, stored as recognition metrics, and categories are associated with each other, similarity of the object to each category included in the recognition table using the second feature value and the plurality of reference feature values;
• outputting the similarity of the object to each category included in the recognition table to an output unit as a result of recognition;
• receiving a category of the object;
• registering the second feature value as the reference feature value in the recognition table with being associated with the category of the object;
• registering the first feature value as the learning sample belonging to the category of the object in the learning sample storage unit;
• and
• performing the generation process again in a case where a predetermined condition is satisfied.

A computer program product comprising a non-transitory computer-readable storage medium having programmed instructions, wherein the instructions, when executed by a computer, cause the computer to execute:

• performing, plural times, a selection process of selecting a plurality of groups each including one or more learning samples from a learning sample storage unit that stores therein a plurality of learning samples that are classified into a plurality of categories;
• learning a classification metric, which is a classifier Fi(x) for each group of the plurality of groups for classifying the plurality of groups selected in each selection process, where 1≦i≦N, and N is the number of classifiers;
• performing a generation process of generating an evaluation metric that includes the classifier Fi(x) for each group of the plurality of groups, wherein the evaluation metric is defined as {F1(x), F2(x), through, FN(x)}, and where x is a variable;
• receiving an image including an object as a recognition target;
• extracting a first feature value representing an image feature of the image;
• transforming the first feature value into a second feature value using the evaluation metric;
• calculating, by referring to a recognition table stored in a recognition table storage unit in which a plurality of reference feature values, stored as recognition metrics, and categories are associated with each other, similarity of the object to each category included in the recognition table using the second feature value and the plurality of reference feature values;
• outputting the similarity of the object to each category included in the recognition table to an output unit as a result of recognition;
• receiving a category of the object;
• registering the second feature value as the reference feature value in the recognition table with being associated with the category of the object;
• registering the first feature value as the learning sample belonging to the category of the object in the learning sample storage unit;
• and
• performing the generation process again in a case where a predetermined condition is satisfied.