Information processing apparatus, method, and program

Patent metadata

Field	Value
Publication number	US-9129149-B2
Application number	US-201113090083-A
Country	US
Kind code	B2
Filing date	Apr 19, 2011
Priority date	Apr 20, 2010
Publication date	Sep 8, 2015
Grant date	Sep 8, 2015

Abstract

Disclosed is a method for accurately and efficiently detecting a modality of input data, including the steps of projecting the input data into a plurality of projection data using each of a plurality of transformation matrix groups U 1 ·(Σ 12 U 2 T ), generating a plurality of inverse projection data by performing inverse projection of the transformation matrix groups on the plurality of generated projection data, calculating a correlation between the input data and the generated inverse projection data with respect to each transformation matrix group U 1 ·(Σ 12 U 2 T ), and identifying a modality represented by a transformation matrix group having a highest calculated correlation as the modality of the input data.

First claim

What is claimed is:

1. An information processing apparatus comprising:
a processor, the processor being programmed to:
learn through each of a plurality of sample data sets and obtain a transformation matrix group including a first projection matrix for projecting input data into a space vector and a second projection matrix for projecting the space vector into projection data having a smaller dimension number than that of the space vector, for each sample data set that has been learned through, wherein each sample data set consists of a plurality of pieces of data which include information in a same state of non-identification target modality among the plurality of pieces of data and different states of identification target modality with respect to each other piece of data among the plurality of pieces of data;
generate a space vector by projecting the input data using the first projection matrix included in the transformation matrix group, for each of the obtained transformation matrix groups;
generate projection data by projecting the space vector using the second projection matrix included in the transformation matrix group, and generate an inverse space vector by performing inverse projection of the second projection matrix on the projection data, for each of the obtained transformation matrix groups;
calculate a correlation between the generated space and the generated inverse space vector, for each of the obtained transformation matrix groups; and
specify a transformation matrix group having a highest calculated correlation among the plurality of transformation matrix groups, and identify the state of identification target modality of the sample data set used in obtaining the specified transformation matrix group, as the state of identification target modality of the input data.

2. The information processing apparatus of claim 1 , wherein each of the transformation matrix group is a multi-linear transformation matrix constituted by a plurality of space transformation matrices, including the first and second projection matrices.

3. The information processing apparatus of claim 1 , wherein the processor is further programmed to perform interpolation process on the input data as a pre-process, and determine a sampling frequency according to the accuracy of the identification of the state of identification target modality.

4. The information processing apparatus of claim 2 , wherein the processor is further programmed to perform interpolation process on the input data, as a pre-processor and determine a sampling frequency according to the accuracy of the identification of the state of identification target modality.

5. An information processing method comprising the steps of:
learning through each of a plurality of sample data sets and obtaining a transformation matrix group including a first projection matrix for projecting input data into a space vector and a second projection matrix for projecting the space vector into projection data having a smaller dimension number than that of the space vector, for each sample data set that has been learned through, wherein each sample data set consists of a plurality of pieces of data which include information in a same state of non-identification target modality among the plurality of pieces of data and different states of identification target modality with respect to each other piece of data among the plurality of pieces of data;
generating a space vector by projecting the input data using the first projection matrix included in the transformation matrix group, for each of the obtained transformation matrix groups;
generating projection data by projecting the space vector using the second projection matrix included in the transformation matrix group, and generating an inverse space vector by performing inverse projection of the second projection matrix on the projection data, for each of the obtained transformation matrix groups;
calculating a correlation between the generated space vector and the generated inverse space vector, for each of the obtained transformation matrix groups; and
specifying a transformation matrix group having a highest calculated correlation among the plurality of transformation matrix groups, and identifying the state of identification target modality of the sample data set used in obtaining the specified transformation matrix group, as the state of identification target modality of the input data.

6. A non-transitory computer readable recording medium on which is recorded an information processing program for causing a computer to perform the steps of:
learning through each of a plurality of sample data sets and obtaining a transformation matrix group including a first projection matrix for projecting input data into a space vector and a second projection matrix for projecting the space vector into projection data having a smaller dimension number than that of the space vector, for each sample data set that has been learned through, wherein each sample data set consists of a plurality of pieces of data which include information in a same state of non-identification target modality among the plurality of pieces of data and different states of identification target modality with respect to each other piece of data among the plurality of pieces of data;
generating a space vector by projecting the input data using the first projection matrix included in the transformation matrix group, for each on the obtained transformation matrix groups;
generating projection data by projecting the space vector using the second projection matrix included in the transformation matrix group, and generating an inverse space vector by performing inverse projection of the second projection matrix on the projection data, for each of the obtained transformation matrix groups;
calculating a correlation between the generated space vector and the generated inverse space vector, for each of the obtained transformation matrix groups; and
specifying a transformation matrix group having a highest calculated correlation among the plurality of transformation matrix groups, and identifying the state of identification target modality of the sample data set used in obtaining the specified transformation matrix group, as the state of identification target modality of the input data.

Assignees

• Kameyama Hirokazu
• Yamaguchi Kouji
• Fujifilm Corp

Inventors

• Kameyama Hirokazu
• Yamaguchi Kouji

Classifications

• G06K9/00275 Primary

  Physics · mapped topic

• G06V40/169 Primary

  Holistic features and representations, i.e. based on the facial image taken as a whole · CPC title