Image processing apparatus, image processing method, image recognition apparatus, and image recognition method

An information processing method, comprising the steps of:

• obtaining a first feature quantity of N dimensions, N being an integer of at least two, from first information prepared for executing learning;
• generating a plurality of first feature quantity combinations having corresponding dimensions that fail to exceed the N dimensions of said first feature quantity;
• obtaining a learning model feature quantity matching each dimension of said first feature quantity combinations;
• computing a first correlation coefficient between said plurality of first feature quantity combinations and said learning model feature quantity;
• and
• classifying predetermined second information by classifying said first information by use of said first correlation coefficient,
• wherein generating the first feature quantity combinations comprises:
• obtaining a matrix comprising elements corresponding to the N dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generating said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A non-transitory computer-readable medium for causing a computer to execute learning for classifying, comprising the steps of:

• obtaining a first feature quantity of N dimensions, N being an integer of at least two, from first information prepared for executing learning;
• generating a plurality of first feature quantity combinations having corresponding dimensions that fail to exceed the N dimensions of said first feature quantity;
• obtaining a learning model feature quantity matching each dimension of said plurality of first feature quantity combinations;
• computing a first correlation coefficient between said plurality of first feature quantity combinations and said learning model feature quantity;
• and
• classifying predetermined second information by classifying said first information by use of said first correlation coefficient,
• wherein generating the first feature quantity combinations comprises:
• obtaining a matrix comprising elements corresponding to the N dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generating said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A recognition apparatus capable of executing recognition processing for classifying predetermined information by use of a predetermined learner, comprising:

• combination generating means for obtaining, from said predetermined information, a feature quantity having less than N dimensions determined beforehand on a basis of said learner to generate a plurality of feature quantity combinations having less than N dimensions;
• and
• classification processing executing means for computing a correlation coefficient between said of feature quantity combinations and a registered feature quantity matching each dimension of said feature quantity combinations to classify said predetermined information on the basis of said learner by use of said correlation coefficient,
• wherein said combination generating means is further configured to:
• obtain a matrix comprising elements corresponding to the dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generate said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

An information recognition method for use in a recognition apparatus capable of obtaining a feature quantity of N dimensions, N being an integer of at least two, to execute recognition processing for classifying predetermined information by use of a predetermined learner, comprising the steps of:

• receiving said predetermined information;
• obtaining, from said predetermined information, a predetermined feature quantity having less than N dimensions determined beforehand on the basis of said learner;
• generating a plurality of feature quantity combinations having corresponding dimensions less than N dimensions of said feature quantity;
• obtaining a registered feature quantity matching each dimension of said plurality of feature quantity combinations;
• and
• computing a correlation coefficient between said plurality of feature quantity combinations and said registered feature quantity to classify said predetermined information by use of said correlation coefficient,
• wherein generating the first feature quantity combinations comprises:
• obtaining a matrix comprising elements corresponding to the N dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generating said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A non-transitory computer-readable medium for causing program configured to make a computer to execute recognition processing capable of obtaining a feature quantity of N dimensions, N being an integer of at least two, to execute recognition processing for classifying predetermined information by use of a predetermined learner, comprising the steps of:

• receiving said predetermined information;
• obtaining, from said predetermined information, a predetermined feature quantity having less than N dimensions determined beforehand on the basis of said learner;
• generating a plurality of feature quantity combinations having corresponding dimensions less than N dimensions of said feature quantity;
• obtaining a registered feature quantity matching each dimension of said plurality of feature quantity combinations;
• and
• computing a correlation coefficient between said plurality of feature quantity combinations and said registered feature quantity to classify said predetermined information by use of said correlation coefficient,
• wherein generating the first feature quantity combinations comprises:
• obtaining a matrix comprising elements corresponding to the N dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generating said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A recognition apparatus capable of obtaining a feature quantity N dimensions, N being an integer of at least two, to execute recognition processing for classifying predetermined information, comprising:

• storage means for storing a predetermined learner obtained as a result of learning processing and information about said feature quantity;
• combination generating means for obtaining, from said information, said feature quantity and generating a plurality of feature quantity combinations having corresponding dimensions that fail to exceed the N dimensions of said feature quantity;
• and
• classification processing executing means for computing a correlation coefficient between said plurality of feature quantity combinations and a registered feature quantity matching each dimension of said plurality of feature quantity combinations, predetermined information being classified on the basis of said learner by use of said correlation coefficient;
• wherein said combination generating means is further configured to:
• obtain a matrix comprising elements corresponding to the N dimensions of the feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generate said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

An information recognition method for use in a recognition apparatus having a predetermined learner and capable executing recognition processing for classifying predetermined information, and the method comprising the steps of:

• obtaining a feature quantity of N dimensions, N being an integer of at least two from a storage;
• generating a plurality of feature quantity combinations having corresponding dimensions that fail to exceed the N dimensions of said feature quantity from said predetermined information by use of said learner and said feature quantity;
• obtaining a registered feature quantity matching each dimension of said plurality of feature quantity combinations;
• computing a correlation coefficient between said generated plurality of feature quantity combinations and said registered feature quantity;
• and
• classifying said predetermined information on the basis of said learner by use of said correlation coefficient,
• wherein generating the feature quantity combinations comprises:
• obtaining a matrix comprising elements corresponding to the N dimensions of the feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generating said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A non-transitory computer-readable medium for causing program configured to make a computer to execute recognition processing for classifying predetermined information by use of a predetermined learner obtained as a result of learning processing, comprising the steps of:

• obtaining a feature quantity of N dimensions, N being an integer of at least two from a storage;
• generating a plurality of feature quantity combinations having corresponding dimensions that fail to exceed the N dimensions of said feature quantity from said predetermined information by use of said learner and said feature quantity;
• obtaining a registered feature quantity matching each dimension of said plurality of feature quantity combinations;
• computing a correlation coefficient between said generated plurality of feature quantity combinations and said registered feature quantity;
• and
• classifying said predetermined information on the basis of said learner by use of said correlation coefficient,
• wherein generating the feature quantity combinations comprises:
• obtaining a matrix comprising elements corresponding to the N dimensions of the feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generating said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

An information processing apparatus, comprising:

• a combination generator configured to obtain a first feature quantity of N dimensions, N being an integer of at least two, from first information prepared for execution of learning and using said first feature quantity to generate at least two first feature quantity having corresponding dimensions that fail to exceed the N dimensions of said first feature quantity;
• and
• a learning processing executer configured to compute a correlation coefficient between said first feature quantity combinations and a learning model feature quantity matching each dimension of said first feature quantity combinations and, by use of said first correlation coefficient, classifying said first information to execute learning processing for classifying predetermined second information,
• wherein said combination generator is further configured to:
• obtain a matrix comprising elements corresponding to the N dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generate said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A recognition apparatus capable of executing recognition processing for classifying predetermined information by use of a predetermined learner obtained as a result of learning processing, comprising:

• a combination generator configured to obtain, from said predetermined information, said predetermined feature quantity having less than N dimensions determined beforehand on a basis of said learner to generate a plurality of feature quantity combinations having less than N dimensions of said feature quantity;
• and
• a classification processing executer configured to compute a correlation coefficient between said feature quantity combinations and a registered feature quantity matching each dimension of said feature quantity combinations to classify said predetermined information on the basis of said learner by use of said correlation coefficient,
• wherein said classification processing executer is further configured to:
• obtain a matrix comprising elements corresponding to the dimensions of the first feature quantity, the elements being output by Gabor filters associated with corresponding ones of the dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generate said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.

A recognition apparatus capable of executing recognition processing for classifying predetermined information, comprising:

• a storage configured to store a predetermined learner obtained as a result of learning processing and information about a feature quantity;
• a combination generator configured to obtain, from said predetermined information, feature quantity having N dimensions, N being an integer of at least two, and generating a plurality of feature quantity combinations having corresponding dimensions that fail to exceed the N dimensions of said feature quantity;
• and
• a classification processing executer configured to compute a correlation coefficient between said plurality of feature quantity combinations and a registered feature quantity matching each dimension of said plurality of feature quantity combinations, predetermined information being classified on the basis of said learner by use of said correlation coefficient;
• wherein said classification processing executer is further configured to:
• obtain a matrix comprising elements corresponding to the N dimensions of the feature quantity, the elements being output by Gabor filters associated with corresponding ones of the N dimensions and being arranged within the matrix according to filter directions and frequencies associated with the Gabor filters;
• and
• generate said first feature quantity combinations by combining elements within corresponding continuous areas of the matrix, the continuous areas being associated with ranges of the filter directions and frequencies of the Gabor filters.