Age estimation apparatus, age estimation method, and age estimation program

• US 8,818,111 B2
• Filed: 04/14/2010
• Issued: 08/26/2014
• Est. Priority Date: 04/28/2009
• Status: Active Grant

• Current Assignee: NEC Solution Innovators, Ltd (NEC Corporation)
• Original Assignee: Tokyo Institute of Technology (National University Corporation Tokyo Institute of Technology), NEC Soft Limited (NEC Corporation)
• Inventors: Ueki, Kazuya, Sugiyama, Masashi, Ihara, Yasuyuki
• Primary Examiner(s): Rahmjoo, Mike
• Application Number:
  US13/263,059
• Publication Number:
  US 20120051629A1
• Time in Patent Office: 1,595 Days
• Field of Search: 382224-231, 382/218, 382/181, 382/182, 382/190, 382/195
• US Class Current: 382/224
• CPC Class Codes: G06F 18/2414 Smoothing the distance, e.g. radial basis function networks [RBFN]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions
• CPC Class Codes: G06V 40/178 estimating age from face image; using age information for improving recognition

Abstract

Provided are an age estimation apparatus, an age estimation method, and an age estimation program capable of reducing the labor of labeling the image data used for age estimation. An age estimation apparatus for estimating an age of a person on image data includes a dimension compression unit for applying dimension compression to the image data to output low dimensional data; a clustering unit for performing clustering of the low dimensional data outputted; a labeling unit for labeling representative data of each cluster among the low dimensional data clustered; and an identification unit for estimating an age of a person on the basis of a learning result using a feature amount contained in labeled low dimensional data and unlabeled low dimensional data.

Claims

1. An age estimation apparatus of estimating an age of an object on image data, comprising:

a dimension compression unit of applying dimension compression to the image data to output low dimensional data;

a clustering unit of performing clustering of the low dimensional data outputted;

a labeling unit that applies a correct age to a part of each clustered low dimensional data to form labeled training data and unlabeling the low dimensional data other than representative data to form unlabeled training data; and

an identification unit for minimizing a value of a loss function to set a learned parameter, wherein

the loss function is a function defined on the basis of an error between an age which is correct in the labeled training data and an age estimated by the identification unit and a term which is increased when a difference between an age estimated from the labeled training data and an age estimated from the unlabeled training data is large even though a distance between the labeled training data and the unlabeled training data is small, and

when predetermined image data, on which an object of age estimation is taken, is inputted, the dimension compression unit applies dimension compression to the image data and outputs the image data as the low dimensional data to the identification unit, and the identification unit estimates an age of the object on the image data on the basis of the dimension compression low dimensional data and the learned parameter.

2. The apparatus according to claim 1, wherein

the loss function is a function defined on the basis of the error, the term and a weighting function which is a function of age for determining an age group in which the error is small.

3. The apparatus according to claim 2, comprising

an evaluation unit for evaluating the age of the object, outputted from the identification unit, as a result of the age estimation on the basis of an error evaluation formula and selecting and setting one or both of parameters with the minimum errors, used in the dimension compression unit and the identification unit, on the basis of a result of the evaluation, wherein

an age evaluation formula is a function defined on the basis of an error between an age which is correct in the labeled training data and an age estimated by the identification unit and a weighting function which is a function of age, for determining an age group in which the error is small.

4. The apparatus according to claim 2, wherein the weighting function is defined on the basis of a standard deviation of each perceptual age.

5. The apparatus according to claim 2, wherein the smaller the value of the weight function is, the more significant the effect of the estimated age error.

6. An age estimation method for estimating an age of an object on image data, comprising:

a dimension compression step of applying dimension compression to the image data to output low dimensional data;

a clustering step of performing clustering of the low dimensional data outputted;

a labeling step that applies a correct age to a part of each clustered low dimensional data to form labeled training data and unlabeling the low dimensional data other than representative data to form unlabeled training data; and

a first setting step for setting a learned parameter learned by minimizing a value of a loss function in an identification device, wherein

the loss function is a function defined on the basis of an error between an age which is correct in the labeled training data and an age estimated by the identification device and a term which is increased when a difference between an age estimated from the labeled training data and an age estimated from the unlabeled training data is large even though a distance between the labeled training data and the unlabeled training data is small, and

when predetermined image data, on which an object of age estimation is taken, is inputted, dimension compression is applied to the image data, the image data is outputted to the identification device as the low dimensional data, and the identification device estimates an age of the object on the image data on the basis of the dimension compression-applied low dimensional data and the learned parameter.

7. The method according to claim 6, wherein

the loss function is a function defined on the basis of the error, the term and a weighting function which is a function of age, for determining an age group in which the error is small.

8. The method according to claim 7, further comprising:

an evaluation step for evaluating the age of the object, outputted from the identification device as a result of the age estimation by an error evaluation formula, and selecting and setting one or both of the parameters with the minimum errors used in the dimension compression and the identification device by the result of the evaluation, wherein

the error evaluation formula is a function defined by an error between an age which is correct in the labeled training data and an age estimated by the identification unit and a weighting function which is a function of age for determining an age group in which the error is small.

9. The method according to claim 7, wherein the weighting function is a function representing a deviation of each perceptual age.

10. The method according to claim 7, wherein

the weighting function is a function representing that the smaller a value of the weighting function is, the more significant the effect of an estimated age error.

11. A non-transitory computer-readable recording medium having an age estimation program that estimates an age of an object on image data recorded, wherein a computer is caused to execute

dimension compression processing of applying dimension compression to the image data to output low dimensional data;

clustering processing of performing clustering of the low dimensional data outputted;

labeling processing that applies a correct age to a part of each clustered low dimensional data to form labeled training data and unlabeling the low dimensional data other than representative data to form unlabeled training data; and

first setting processing of setting a learned parameter learned by minimizing a value of a loss function, wherein

the loss function is a function defined on the basis of an error between an age which is correct in the labeled training data and an age estimated by the identification unit and a term which is increased when a difference between an age estimated from the labeled training data and an age estimated from the unlabeled training data is large even though a distance between the labeled training data and the unlabeled training data is small, and

when predetermined image data, on which an object of age estimation is taken, is inputted, the computer is caused to apply dimension compression to the image data to form the low dimensional data and execute age estimation processing of estimating an age of the object on the image data on the basis of the low dimensional data and the learned parameter.

12. The computer-readable recording medium according to claim 11, wherein

the loss function is a function defined on the basis of the error, the term and a weighting function which is a function of age, for determining an age group in which the error is small.

13. The computer-readable recording medium according to claim 12, wherein the computer is further caused to execute:

an evaluating processing of evaluating the age of the object, outputted as a result of the age estimation by an error evaluation formula and selecting and setting one or both of parameters with the minimum errors used in the dimension compression and the age estimation by a result of the evaluation, and

an age evaluation formula is a function defined on the basis of an error between an age which is correct in the labeled training data and an age estimated and a term which is increased when a difference between an age estimated from the labeled training data and an age estimated from the unlabeled training data is large even though a distance between the labeled training data and the unlabeled training data is small.

14. The computer-readable recording medium according to claim 12, wherein the computer is caused to execute processing of defining the weighting function on the basis of a standard deviation of each perceptual age.

15. The computer-readable recording medium according to claim 12, wherein the weighting function is a function representing that the smaller the value of the weighting function is, the more serious the estimated age error.