Image processing method, image processing apparatus, and computer readable medium, in which an image processing program is recorded
AssigneeFUJIFILM Corporation
InventorYuanzhong Li
Life of this patent
Estimating the term…
Abstract
A plurality of landmarks, which are employed in statistical models, such as an ASM, that is, a plurality of landmarks that represent the shape and the like of a predetermined subject, such as a face, within an image, are detected. Discrimination regarding whether each pixel within the image is a point that represents a landmark is performed based on discrimination conditions. The discrimination conditions correspond to characteristic amounts, which are obtained by learning characteristic amounts of positions, which are known to be landmarks, and characteristic amounts of positions, which are known to not be landmarks, within a plurality of sample images of the subject, by a machine learning technique.
Claims19
An image processing method for detecting the positions of a plurality of landmarks that indicate the shape of a predetermined subject by their positions and/or their positional relationships, from within the predetermined subject included in an image, comprising the steps of:
- setting the positions of a plurality of landmarks that indicate the average shape of the predetermined subject as preliminary positions of the landmarks included in the predetermined subject included in the image;
- calculating characteristic amounts, which are defined for a landmark corresponding to a preliminary position, of each pixel within a range of pixels that includes the preliminary position to discriminate the landmark, wherein the calculating characteristic amounts is performed by a processor;
- judging whether pixels that represent the landmark are included in the range, by discriminating whether each of the pixels within the range represents the landmark;
- moving the preliminary position to approach the position of the pixel which has been discriminated to represent the landmark, in the case that the judgment is affirmative;
- repeating the preceding three steps for all of the preliminary positions;
- and
- determining the positions of the preliminary positions after they have been moved to be the positions of the landmarks that correspond to each of the preliminary positions;
- the discrimination regarding whether the pixels represent the landmarks being performed, based on learning characteristic amounts of positions, which are known to represent the landmarks, and characteristic amounts of positions, which are known to not represent the landmarks, of a plurality of sample images by a machine learning technique.
An image processing apparatus for detecting the positions of a plurality of landmarks that indicate the shape of a predetermined subject by their positions and/or their positional relationships, from within the predetermined subject included in an image, comprising:
- preliminary position setting means, for setting the positions of a plurality of landmarks that indicate the average shape of the predetermined subject as preliminary positions of the landmarks included in the predetermined subject included in the image;
- moving means, for:
- calculating characteristic amounts, which are defined for a landmark corresponding to a preliminary position, of each pixel within a range of pixels that includes the preliminary position to discriminate the landmark;
- judging whether pixels that represent the landmark are included in the range, by discriminating whether each of the pixels within the range represents the landmark;
- moving the preliminary position to approach the position of the pixel which has been discriminated to represent the landmark, in the case that the judgment is affirmative;
- and repeating the preceding three steps for all of the preliminary positions;
- and
- landmark position determining means, for determining the positions of the preliminary positions after they have been moved to be the positions of the landmarks that correspond to each of the preliminary positions;
- the moving means performing discrimination regarding whether the pixels represent the landmarks, based on learning characteristic amounts of positions, which are known to represent the landmarks, and characteristic amounts of positions, which are known to not represent the landmarks, of a plurality of sample images by a machine learning technique.
A computer readable recording medium having recorded therein a program that causes a computer to execute an image processing method for detecting the positions of a plurality of landmarks that indicate the shape of a predetermined subject by their positions and/or their positional relationships, from within the predetermined subject included in an image, comprising the procedures of:
- setting the positions of a plurality of landmarks that indicate the average shape of the predetermined subject as preliminary positions of the landmarks included in the predetermined subject included in the image;
- calculating characteristic amounts, which are defined for a landmark corresponding to a preliminary position, of each pixel within a range of pixels that includes the preliminary position to discriminate the landmark;
- judging whether pixels that represent the landmark are included in the range, by discriminating whether each of the pixels within the range represents the landmark;
- moving the preliminary position to approach the position of the pixel which has been discriminated to represent the landmark, in the case that the judgment is affirmative;
- repeating the preceding three steps for all of the preliminary positions;
- and
- determining the positions of the preliminary positions after they have been moved to be the positions of the landmarks that correspond to each of the preliminary positions;
- the discrimination regarding whether the pixels represent the landmarks being performed, based on learning characteristic amounts of positions, which are known to represent the landmarks, and characteristic amounts of positions, which are known to not represent the landmarks, of a plurality of sample images by a machine learning technique.
Dependencies are inferred from the claim text.