Image processing method, image processing apparatus, and computer program product

1. An image processing method for extracting a region having a specific feature, the image processing method comprising:

extracting pixels representing an object, to which the pixel having the specific feature belongs, from a given input image based on an image feature other than color;

extracting principal colors from the pixels;

selecting, from a plurality of pre-trained color distribution models deposited in a repository, a color distribution model closest in a distance metric to the extracted principal colors by performing a search through the repository using the extracted principal colors, the distance metric indicating distance relative to the extracted principal colors;

segmenting the input image by categorizing pixels in the input image using the selected color distribution model; and

modifying a Gaussian mixture model (hereinafter, “GMM”) obtained from the input image based on score values of components of the GMM obtained from the input image, wherein in the step of modifying:

a score value is an output obtained by inputting a mean of the components of the GMM of the input image to the GMM deposited in the repository,

the score value of each of the components of the GMM of the input image is compared with a predetermined threshold value,

any component of which score value is equal to or lower than the threshold value of the components is deleted from the GMM of the input image, and

weights of remaining components, the remaining components being the components of the GMM of the input image exclusive of the deleted component, are re-defined so that a sum of the weights of the remaining components is one.

2. The image processing method according to claim 1 , wherein

the object, to which the pixel having the specific feature belongs, is a human face, and

the step of extracting pixels includes extracting a face region.

3. The image processing method according to claim 1 , wherein the step of extracting principal colors includes:

calculating a probability density distribution from distribution of the pixels in a predetermined color space, and

approximating the probability density distribution by the GMM.

4. The image processing method according to claim 1 , wherein the distance metric is an inverse of probability density distribution obtained as an output value by inputting the plurality of principal colors extracted from the input image to the color distribution model deposited in the repository.

5. The image processing method according to claim 1 , further comprising segmenting the input image by categorizing the pixels in the input image using the modified GMM when the search results in failure.

6. An image processing apparatus for extracting a region having a specific feature, the image processing apparatus comprising:

circuitry configured to:

extract pixels representing an object, to which the pixel having the specific feature belongs, from a given input image based on an image feature other than color;

extract principal colors from the pixels;

select, from a plurality of pre-trained color distribution models deposited in a repository, a color distribution model closest in a distance metric to the extracted principal colors by performing a search through the repository using the extracted principal colors, the distance metric indicating distance relative to the extracted principal colors;

segment the input image by categorizing pixels in the input image using the selected color distribution model; and

modify a Gaussian mixture model (hereinafter, “GMM”) obtained from the input image based on score values of components of the GMM obtained from the input image, wherein

a score value is an output obtained by inputting a mean of the components of the GMM of the input image to the GMM deposited in the repository,

the score value of each of the components of the GMM of the input image is compared with a predetermined threshold value,

any component of which score value is equal to or lower than the threshold value of the components is deleted from the GMM of the input image, and

weights of remaining components, the remaining components being the components of the GMM of the input image exclusive of the deleted component, are re-defined so that a sum of the weights of the remaining components is one.

7. A non-transitory computer-readable recording medium that contains a computer program that when executed causes a computer to:

extract pixels representing an object, to which the pixel having the specific feature belongs, from a given input image based on an image feature other than color;

extract principal colors from the pixels;

select, from a plurality of pre-trained color distribution models deposited in a repository, a color distribution model closest in a distance metric to the extracted principal colors by performing a search through the repository using the extracted principal colors, the distance metric indicating distance relative to the extracted principal colors;

segment the input image by categorizing pixels in the input image using the selected color distribution model; and

modify a Gaussian mixture model (hereinafter, “GMM”) obtained from the input image based on score values of components of the GMM obtained from the input image, wherein

a score value is an output obtained by inputting a mean of the components of the GMM of the input image to the GMM deposited in the repository,

the score value of each of the components of the GMM of the input image is compared with a predetermined threshold value,

any component of which score value is equal to or lower than the threshold value of the components is deleted from the GMM of the input image, and

weights of remaining components, the remaining components being the components of the GMM of the input image exclusive of the deleted component, are re-defined so that a sum of the weights of the remaining components is one.