Data converting apparatus, data converting method, learning apparatus, leaning method, program, and recording medium

A data conversion method executed by a processor for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• an inputting step for inputting the first data;
• a class-tap generating step of generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• a classification step of classifying the specified sample based on the class tap;
• a predictive-tap generating step of generating, from the first data, a predictive tap for determining the specified sample;
• a main tap-coefficient obtaining step for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• a sub tap-coefficient obtaining step of obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between supervisor data corresponding to the second data, which serves as a learning supervisor, and learner data corresponding to the first data, which serves as a learner, for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• a computation step of determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• an outputting step for outputting the second data composed of samples computed by the computation step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A computer readable medium storing a program including program in instructions for, when executed by a processor, allowing a computer to execute data conversion processing for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• an inputting step for inputting the first data;
• a class-tap generating step of generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• a classification step of classifying the specified sample based on the class tap;
• a predictive-tap generating step of generating, from the first data, a predictive tap for determining the specified sample;
• a main tap-coefficient obtaining step for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• a sub tap-coefficient obtaining step of obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between supervisor data corresponding to the second data, which serves as a learning supervisor, and learner data corresponding to the first data, which serves as a learner, for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• a computation step of determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• an outputting step for outputting the second data composed of samples computed by the computation step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A recording medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute data conversion processing for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data is recorded, said program comprising:

• an inputting step for inputting the first data;
• a class-tap generating step of generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• a classification step of classifying the specified sample based on the class tap;
• a predictive-tap generating step of generating, from the first data, a predictive tap for determining the specified sample;
• a main tap-coefficient obtaining step for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• a sub tap-coefficient obtaining step of obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between supervisor data corresponding to the second data, which serves as a learning supervisor, and learner data corresponding to the first data, which serves as a learner, for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• a computation step of determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• an outputting step for outputting the second data composed of samples computed by the computation step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A learning device for performing learning for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• inputting means for inputting the first data;
• class-tap generating means for generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• classification means for classifying the specified item of data based on the class tap;
• predictive-tap generating means for generating a predictive tap used for determining the specified item of data from the learner data;
• learning means for determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the sub tap coefficient for each of said at least one class by learning a relationship between the supervisor data and the learner data for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• and
• outputting means for outputting the second data composed of samples computed by the learning means,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A learning method executed by a processor for performing learning for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• inputting means for inputting the first data;
• a class-tap generating step of generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• a classification step of classifying the specified item of data based on the class tap;
• a predictive-tap generating step of generating a predictive tap used for determining the specified item of data from the learner data;
• a learning step of determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the sub tap coefficient for each of said at least one class by learning a relationship between the supervisor data and the learner data for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• and
• an outputting step for outputting the second data composed of samples computed by the learning step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A computer readable medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute learning processing for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• an inputting step for inputting the first data;
• a class-tap generating step of generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• a classification step of classifying the specified item of data based on the class tap;
• a predictive-tap generating step of generating a predictive tap used for determining the specified item of data from the learner data;
• a learning step of determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the main tap coefficient for each of said at least one class by learning a relationship between the supervisor data and the learner data for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• and
• an outputting step for outputting the second data composed of samples computed by the learning step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A recording medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute learning processing for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, said program comprising:

• a class-tap generating step of generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• inputting the first data;
• a classification step of classifying the specified item of data based on the class tap;
• a predictive-tap generating step of generating a predictive tap used for determining the specified item of data from the learner data;
• a learning step of determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the main tap coefficient for each of said at least one class by learning a relationship between the supervisor data and the learner data for each of said at least one class by giving a predetermined constraint condition to the supervisor data;
• and
• an outputting step for outputting the second data composed of samples computed by the learning step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the predetermined constraint condition represents a difference between the main sample and the sub sample of the second data.

A data conversion apparatus for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• class-tap generating means for generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• inputting means for inputting the first data;
• classification means for classifying the specified sample based on the class tap;
• predictive-tap generating means for generating, from the first data, a predictive tap for determining the specified sample;
• main tap-coefficient obtaining means for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• sub tap-coefficient obtaining means for obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between a feature obtained from a plurality of samples of supervisor data corresponding to the second data, which serves as a learning supervisor, and a plurality of samples of learner data corresponding to the first data, which serves as a learner, for each of said at least one class;
• computation means for determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• outputting means for outputting the second data composed of samples computed by the computation means,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A data conversion method executed by a processor for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• a class-tap generating step of generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• an inputting step for inputting the first data;
• a classification step of classifying the specified sample based on the class tap;
• a predictive-tap generating step of generating, from the first data, a predictive tap for determining the specified sample;
• a main tap-coefficient obtaining step for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• a sub tap-coefficient obtaining step of obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between a feature obtained from a plurality of samples of supervisor data corresponding to the second data, which serves as a learning supervisor, and a plurality of samples of learner data corresponding to the first data, which serves as a leaner, for each of said at least one class;
• a computation step of determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• an outputting step for outputting the second data composed of samples computed by the computation step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A computer readable medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute data conversion processing for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• a class-tap generating step of generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• an inputting step for inputting the first data;
• a classification step of classifying the specified sample based on the class tap;
• a predictive-tap generating step of generating, from the first data, a predictive tap for determining the specified sample;
• a main tap-coefficient obtaining step for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• a sub tap-coefficient obtaining step of obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between a feature obtained from a plurality of samples of supervisor data corresponding to the second data, which serves as a learning supervisor, and a plurality of samples of learner data corresponding to the first data, which serves as a learner, for each of said at least one class;
• a computation step of determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• an outputting step for outputting the second data composed of samples computed by the computation step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A recording medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute data conversion processing for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data is recorded, said program comprising:

• a class-tap generating step of generating, from the first data, a class tap used for classifying a specified sample being a main sample or a sub sample of the second data into a class of at least one class;
• an inputting step for inputting the first data;
• a classification step of classifying the specified sample based on the class tap;
• a predictive-tap generating step of generating, from the first data, a predictive tap for determining the specified sample;
• a main tap-coefficient obtaining step for obtaining a main tap coefficient for the class of the specified sample from tap coefficients obtained by learning for each of said at least one class;
• a sub tap-coefficient obtaining step of obtaining a sub tap coefficient for the class of the specified sample from tap coefficients obtained by learning a relationship between a feature obtained from a plurality of samples of supervisor data corresponding to the second data, which serves as a learning supervisor, and a plurality of samples of learner data corresponding to the first data, which serves as a learner, for each of said at least one class;
• a computation step of determining the specified sample by using the predictive tap and the main tap coefficient and the sub tap coefficient for the class of the specified sample;
• and
• an outputting step for outputting the second data composed of samples computed by the computation step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A learning device for performing learning for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• class-tap generating means for generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• inputting means for inputting the first data;
• classification means for classifying the specified item of data based on the class tap;
• predictive-tap generating means for generating a predictive tap used for determining the specified item of data from the learner data;
• learning means for determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the sub tap coefficient for each of said at least one class by learning a relationship between a feature obtained from a plurality of samples of the supervisor data and a plurality of samples of the learner data for each of said at least one class;
• an outputting means for outputting the second data composed of samples computed by the learning means,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A learning method executed by a processor for performing learning for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• a class-tap generating step of generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• an inputting step for inputting the first data;
• a classification step of classifying the specified item of data based on the class tap;
• a predictive-tap generating step of generating a predictive tap used for determining the specified item of data from the learner data;
• a learning step of determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the sub tap coefficient for each of said at least one class by learning a relationship between a feature obtained from a plurality of samples of the supervisor data and a plurality of samples of the learner data for each of said at least one class;
• and
• an outputting step for outputting the second data composed of samples computed by the learning step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A computer readable medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute learning processing for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data, comprising:

• a class-tap generating step of generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• an inputting step for inputting the first data;
• a classification step of classifying the specified item of data based on the class tap;
• a predictive-tap generating step of generating a predictive tap used for determining the specified item of data from the learner data;
• a learning step of determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the sub tap coefficient for each of said at least one class by learning a relationship between a feature obtained from a plurality of samples of the supervisor data and a plurality of samples of the learner data for each of said at least one class;
• and
• an outputting step for outputting the second data composed of samples computed by the learning step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.

A recording medium storing a program including program instructions for, when executed by a processor, allowing a computer to execute learning processing for determining a predetermined tap coefficient used for producing high definition image data by converting standard definition image data, inputted as first data, into high definition image data, outputted as second data is recorded, said program comprising:

• a class-tap generating step of generating a class tap used for classifying a specified item being a main sample or a sub sample of supervisor data corresponding to the second data, which serves as a supervisor for learning a main tap coefficient and a sub tap coefficient, into a class of at least one class, from learner data corresponding to the first data, which serves as a learner;
• an inputting step for inputting the first data;
• a classification step of classifying the specified item of data based on the class tap;
• a predictive-tap generating step of generating a predictive tap used for determining the specified item of data from the learner data;
• a learning step of determining, by using the specified item of data and the predictive tap, the main tap-coefficient for the class of the specified item and for determining the sub tap coefficient for each of said at least one class by learning a relationship between a feature obtained from a plurality of samples of the supervisor data and a plurality of samples of the learner data for each of said at least one class;
• and
• an outputting step for outputting the second data composed of samples computed by the learning step,
• wherein the main tap coefficient corresponds to the main sample and the sub tap coefficient corresponds to the sub sample, and
• wherein the feature represents a difference between the main sample and the sub sample of the second data.