Human skin detection based on human-body prior

1. An electronic device, comprising:

circuitry configured to:

acquire a color image of a person;

estimate a three-dimensional (3D) body model of the person based on the acquired color image;

identify one or more unclothed parts of the estimated 3D body model, wherein

the identified one or more unclothed parts correspond to one or more body parts, of which at least a portion of skin remains uncovered by clothes,

the identification includes an assignment of an identifier to each of the one or more unclothed parts of the estimated 3D body model, and

the assignment of the identifier is based on a selection of a corresponding identifier from a set of identifiers associated with a 3D human-body prior;

extract, from the acquired color image, pixel information which corresponds to the identified one or more unclothed parts;

determine, based on the extracted pixel information, classification information comprising a set of values, each of which indicates a likelihood of whether or not a corresponding pixel of the acquired color image is part of an unclothed skin portion of the person; and

detect the unclothed skin portion in the acquired color image based on the determined classification information.

2. The electronic device according to claim 1 , wherein

the circuitry is further configured to acquire depth information associated with a body of the person, and

the estimation of the 3D body model is further based on the acquired depth information.

3. The electronic device according to claim 1 , wherein the circuitry is further configured to:

determine a two-dimensional (2D) pose of a body of the person based on the acquired color image; and

determine a 3D pose of the body based on the determined 2D pose and the 3D human-body prior, wherein

the 3D body model is estimated based on the determined 3D pose and the 3D human-body prior.

4. The electronic device according to claim 1 , wherein

the circuitry is further configured to receive a user input, and

the one or more unclothed parts are identified based on the received user input.

5. The electronic device according to claim 1 , wherein the identified one or more unclothed parts comprises one or more of a face portion, a head portion, a neck portion, a hand portion, or a leg portion.

6. The electronic device according to claim 1 , wherein the circuitry is further configured to:

project the estimated 3D body model onto a 2D image plane; and

extract the pixel information based on a correspondence between points belonging to the identified one or more unclothed parts in the projection of the 3D body model and corresponding pixels of the acquired color image.

7. The electronic device according to claim 1 , wherein the circuitry is further configured to:

determine a first part of the estimated 3D body model based on a color distribution of pixels which are included in the acquired color image and correspond to the first part of the estimated 3D body model, wherein the first part is clothed; and

identify the one or more unclothed parts further based on the determination of the first part.

8. The electronic device according to claim 1 , wherein the circuitry is further configured to:

fit the extracted pixel information to a Gaussian model; and

estimate a mean and covariance of pixel values included in the extracted pixel information, based on the fitting.

9. The electronic device according to claim 8 , wherein the classification information is a probability map which is determined based on the estimated mean and covariance.

10. The electronic device according to claim 9 , wherein

the circuitry is further configured to apply a thresholding operation on the probability map, and

the unclothed skin portion in the acquired color image is detected based on the application of the thresholding operation.

11. A method, comprising:

in an electronic device:

acquiring a color image of a person;

estimating a three-dimensional (3D) body model of the person based on the acquired color image;

identifying one or more unclothed parts of the estimated 3D body model, wherein

the identified one or more unclothed parts correspond to one or more body parts, of which at least a portion of skin remains uncovered by clothes,

the identification includes an assignment of an identifier to each of the one or more unclothed parts of the estimated 3D body model, and

the assignment of the identifier is based on a selection of a corresponding identifier from a set of identifiers associated with a 3D human-body prior;

extracting, from the acquired color image, pixel information which corresponds to the identified one or more unclothed parts;

determining, based on the extracted pixel information, classification information comprising a set of values, each of which indicates a likelihood of whether or not a corresponding pixel of the acquired color image is part of an unclothed skin portion of the person; and

detecting the unclothed skin portion in the acquired color image based on the determined classification information.

12. The method according to claim 11 , further comprising:

determining a two-dimensional (2D) pose of a body of the person based on the acquired color image; and

determining a 3D pose of the body based on the determined 2D pose and the 3D human-body prior, wherein

the 3D body model is estimated based on the determined 3D pose and the 3D human-body prior.

13. The method according to claim 11 , further comprising receiving a user input, wherein the one or more unclothed parts are identified based on the received user input.

14. The method according to claim 11 , further comprising:

projecting the estimated 3D body model onto a 2D image plane; and

extracting the pixel information based on a correspondence between points belonging to the identified one or more unclothed parts in the projection of the 3D body model and corresponding pixels of the acquired color image.

15. The method according to claim 11 , further comprising:

determining a first part of the estimated 3D body model based on a color distribution of pixels which are included in the acquired color image and correspond to the first part of the estimated 3D body model, wherein the first part is clothed; and

identifying the one or more unclothed parts further based on the determination of the first part.

16. The method according to claim 11 , further comprising:

fitting the extracted pixel information to a Gaussian model; and

estimating a mean and covariance of pixel values included in the extracted pixel information, based on the fitting, wherein the classification information is a probability map which is determined based on the estimated mean and covariance.

17. The method according to claim 16 , further comprising applying a thresholding operation on the probability map, wherein the unclothed skin portion in the acquired color image is detected based on the application of the thresholding operation.

18. A non-transitory computer-readable medium having stored thereon, computer-executable instructions that when executed by an electronic device, causes the electronic device to execute operations, the operations comprising:

acquiring a color image of a person;

estimating a three-dimensional (3D) body model of the person based on the acquired color image;

identifying one or more unclothed parts of the estimated 3D body model, wherein

the identified one or more unclothed parts correspond to one or more body parts, of which at least a portion of skin remains uncovered by clothes,

the identification includes an assignment of an identifier to each of the one or more unclothed parts of the estimated 3D body model, and

the assignment of the identifier is based on a selection of a corresponding identifier from a set of identifiers associated with a 3D human-body prior;

extracting, from the acquired color image, pixel information which corresponds to the identified one or more unclothed parts;

determining, based on the extracted pixel information, classification information comprising a set of values, each of which indicates a likelihood of whether or not a corresponding pixel of the acquired color image is part of an unclothed skin portion of the person; and

detecting the unclothed skin portion in the acquired color image based on the determined classification information.

19. An electronic device, comprising:

circuitry configured to:

acquire a color image of a person;

determine a two-dimensional (2D) pose of a body of the person based on the acquired color image;

determine a 3D pose of the body based on the determined 2D pose and a 3D human-body prior;

estimate a three-dimensional (3D) body model of the person based on the determined 3D pose and the 3D human-body prior;

identify one or more unclothed parts of the estimated 3D body model, wherein the identified one or more unclothed parts correspond to one or more body parts, of which at least a portion of skin remains uncovered by clothes;

extract, from the acquired color image, pixel information which corresponds to the identified one or more unclothed parts;

determine, based on the extracted pixel information, classification information comprising a set of values, each of which indicates a likelihood of whether or not a corresponding pixel of the acquired color image is part of an unclothed skin portion of the person; and

detect the unclothed skin portion in the acquired color image based on the determined classification information.