Data processing device, data processing method, and computer program product

A data processing method comprising:

• generating, from a feature vector set including N number of D-dimensional feature vectors, M number of sub-vector groups (where M<D holds true), each of the M number of sub vector groups including N number of dimension-variable sub-vectors obtained from the N number of D-dimensional feature vectors, each of the N number of dimension-variable sub-vectors having, as elements, values of one or more dimensions extracted from the D-dimensional feature vectors, and a number of elements in the dimension-variable sub-vectors in at least one sub-vector group from among the M number of sub-vector groups being different from a number of elements of the dimension-variable sub-vectors in other sub-vector groups;
• wherein the sub-vector group generating unit generates the M number of sub-vector groups in such a way that a degree of dispersion of the dimension-variable sub-vectors is similar among the M number of sub-vector groups;
• performing clustering of the N number of dimension-variable sub-vectors for each of the M number of sub-vector groups, and generating a codebook in which representative vector of each cluster is associated with an index;
• and
• performing product quantization using the codebook and converting each of the N number of D-dimensional feature vectors into a compressed code made of combination of M number of indexes.

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

• generating, from a feature vector set including N number of D-dimensional feature vectors, M number of sub-vector groups (where M<D holds true), each of the M number of sub-vector groups including N number of dimension-variable sub-vectors obtained from the N number of D-dimensional feature vectors, each of the N number of dimension-variable sub-vectors having, as elements, values of one or more dimensions extracted from the D-dimensional feature vectors, and a number of elements in the dimension-variable sub-vectors in at least one sub-vector group from among the M number of sub-vector groups being different from a number of elements of the dimension-variable sub-vectors in other sub-vector groups;
• wherein the sub-vector group generating unit generates the M number of sub-vector groups in such a way that a degree of dispersion of the dimension-variable sub-vectors is similar among the M number of sub-vector groups;
• performing clustering of the N number of dimension-variable sub-vectors for each of the M number of sub-vector groups, and generating a codebook in which a representative vector of each cluster is associated with an index;
• and
• performing product quantization using the codebook and converting each of the N number of D-dimensional feature vectors into a compressed code made of combination of M number of indexes.