Ear shape analysis method, ear shape analysis device, and ear shape model generation method

1. An ear shape analysis method implemented by a computer, the method comprising:

generating a first ear shape data set indicating a difference between a three-dimensional shape of an ear and a three-dimensional shape of a reference ear by applying a first principal component weight vector to an ear shape model reflecting statistical tendencies of three-dimensional shapes of ears; and

identifying from the generated first ear shape data set an estimated three-dimensional shape of a target ear corresponding to a target ear image represented by image data.

2. The ear shape analysis method according to claim 1 , wherein

the ear shape model indicates a relation between second ear shape data sets and second principal component weight vectors, each second ear shape data set indicating a difference between a point group representing a three-dimensional shape of an ear and a point group representing a three-dimensional shape of a reference ear, and each second principal component weight vector indicating weights of principal components of the corresponding second ear shape data set.

3. The ear shape analysis method according to claim 2 , wherein

the generated first ear shape data set is one of a plurality of first ear shape data sets, each corresponding to one of a plurality of candidate ears, and

the generating the first ear shape data set includes applying to the ear shape model each of a plurality of first principal component weight vectors including the first principal component weight vector, to generate each of the plurality of first ear shape data sets, and

the identifying the estimated three-dimensional shape includes,

generating for the plurality of candidate ears a plurality of candidate ear images, each representing a corresponding candidate ear in accordance with the point group representing the three-dimensional shape of the reference ear and the first ear shape data set of the candidate ear, and

comparing the target ear image represented by the image data with each of the plurality of candidate ear images generated for the plurality of candidate ears, to identify as the estimated three-dimensional shape of the target ear an ear shape that corresponds to a candidate ear corresponding to a candidate ear image that has the smallest difference among differences existing between the target ear image and the respective candidate ear images, from among the plurality of candidate ear images.

4. The ear shape analysis method according to claim 3 , wherein

the generating each candidate ear image includes generating a candidate ear image of each candidate ear observed from a viewpoint conforming to conditions close to conditions used when the target ear represented by the image data was captured.

5. The ear shape analysis method according to claim 3 , wherein the generating the first ear shape data set includes,

applying each of the first principal component weight vectors to the ear shape model, to generate the first ear shape data set of each candidate ear, the first ear shape data set including a plurality of translation vectors corresponding to respective points constituting a first group that is a part of the point group of the reference ear, and by interpolation of the plurality of translation vectors included in the first ear shape data set of each candidate ear, generating translation vectors corresponding to respective points constituting a second group of the point group of the reference ear, the second group being constituted by all points of the point group of the reference ear other than the points constituting the first group, and

the generating each candidate ear image includes generating each candidate ear image by moving each of the points constituting the first group of the point group of the reference ear in accordance with a corresponding one of the plurality of translation vectors of the first ear shape data set of the candidate ear, and by moving each of the points constituting the second group of the point group of the reference ear in accordance with a corresponding one of the translation vectors generated by the interpolation.

6. The ear shape analysis method according to claim 2 , further comprising:

generating a principal component weight vector by applying the target ear image represented by the image data to a neural network indicating a relation between ear images and principal component weight vectors, wherein

the generating the first ear shape data set includes generating a first ear shape data set of the target ear by applying the principal component weight vector generated by the neural network to the ear shape model, and

the identifying the estimated three-dimensional shape includes identifying the estimated three-dimensional shape of the target ear in accordance with the point group representing the three-dimensional shape of the reference ear and the first ear shape data set of the target ear.

7. The ear shape analysis method according to claim 1 , further comprising calculating a head-related transfer function corresponding to the estimated three-dimensional shape.

8. The ear shape analysis method according to claim 7 , further comprising receiving the image data from a terminal device, and transmitting to the terminal device the head-related transfer function calculated from the image data.

9. An ear shape analysis device, comprising

an ear shape data generator configured to generate a first ear shape data set indicating a difference between a three-dimensional shape of an ear and a three-dimensional shape of a reference ear by applying a first principal component weight vector to an ear shape model reflecting statistical tendencies of three-dimensional shapes of ears; and

an ear shape identifier configured to identify, from the first ear shape data set generated by the ear shape data generator, an estimated three-dimensional shape of a target ear corresponding to a target ear image represented by image data.

10. The ear shape analysis device according to claim 9 , wherein

the ear shape model indicates a relation between second ear shape data sets and second principal component weight vectors, each second ear shape data set indicating a difference between a point group representing a three-dimensional shape of an ear and a point group representing a three-dimensional shape of a reference ear, and each second principal component weight vector indicating weights of principal components of the corresponding second ear shape data set.

11. The ear shape analysis device according to claim 10 , wherein

the generated first ear shape data set is one of a plurality of first ear shape data sets, each corresponding to one of a plurality of candidate ears, and

the ear shape data generator applies to the ear shape model each of a plurality of first principal component weight vectors including the first principal component weight vector, to generate each of the first ear shape data sets for the plurality of candidate ears, and

the ear shape identifier includes,

an image generator configured to generate for the plurality of candidate ears a plurality of candidate ear images, each representing a corresponding candidate ear, in accordance with the point group representing the three-dimensional shape of the reference ear and the first ear shape data set of the candidate ear, and

an image searcher configured to compare the target ear image represented by the image data with each of the plurality of candidate ear images generated for the plurality of candidate ears, to identify as the estimated three-dimensional shape of the target ear an ear shape that corresponds to a candidate ear corresponding to a candidate ear image that has the smallest difference among differences existing between the target ear image and the respective candidate ear images, from among the plurality of candidate ear images.

12. The ear shape analysis device according to claim 10 , further comprising:

an estimation processor configured to generate a principal component weight vector by applying the target ear image represented by the image data to a neural network indicating a relation between ear images and principal component weight vectors, wherein

the ear shape data generator generates a first ear shape data set of the target ear by applying the principal component weight vector generated by the neural network to the ear shape model, and

the ear shape identifier identifies the estimated three-dimensional shape of the target ear in accordance with the point group representing the three-dimensional shape of the reference ear and the first ear shape data set of the target ear.

13. An ear shape model generation method implemented by a computer to generate an ear shape model used in generating a candidate ear image for comparison with a target ear image represented by image data, the method comprising:

generating, for a plurality of sample ears, a plurality of ear shape data sets, each indicating a difference between a point group representing a three-dimensional shape of a corresponding sample ear and a point group representing a three-dimensional shape of a reference ear, and

calculating a transformation matrix for translating ear shape data into a principal component weight vector indicating weights of principal components, by performing principal component analysis on the plurality of generated ear shape data sets, to generate the ear shape model including the transformation matrix or an inverse matrix of the transformation matrix.

14. The ear shape model generation method according to claim 13 , wherein

the generating the ear shape model includes calculating, for the plurality of sample ears, a mean vector of the ear shape data sets, and generating the ear shape model including the mean vector.

15. The ear shape model generation method according to claim 13 , wherein

the generating the ear shape model includes removing prescribed rows from lower rows of the transformation matrix calculated by the principal component analysis, and generating the ear shape model indicating a transformation matrix resulting from the removing of the prescribed rows.