Method for correcting the position of the eyes in an image
AssigneeThales
InventorJohan D'Hose
Life of this patent
Estimating the term…
Abstract
A method for determining the position of the eyes includes: a step for determining approximate positions of the eyes in the digital image; a step for determining adjacent positions; a step of geometric transformation of the digital image into normalized images in such a way as to make the adjacent positions coincide with the reference positions; a step of comparison of at least a part of each normalized image with a corresponding part of a predetermined image, each comparison providing a score representing the level of resemblance between the parts of images; and a step of declaration of the adjacent positions associated with the best score as actual positions of the eyes of the face.
Claims9
A method for determining positions of eyes of a face in a two-dimensional digital image, spatial dimensions of the two-dimensional digital image consisting of a first spatial dimension and a second spatial dimension, the first spatial dimension being orthogonal to the second spatial dimension, the method comprising:
- determining a pair of approximate positions of the eyes in the two-dimensional digital image;
- determining test pairs of the positions of the eyes on a basis of the pair of approximate positions, each position of an eye in each of the test pairs being in an adjacent area of the approximate position of a corresponding eye;
- geometrically transforming, for each test pair, the two-dimensional digital image into a normalized image in such a way as to make the positions of the eyes of each test pair coincide with a pair of predetermined reference positions;
- comparing at least a part of each normalized image covering at least a part of the face with a corresponding part of a predetermined image, each comparison providing a score representing a level of resemblance between the part of the normalized image and that of the predetermined image;
- and
- declaring the positions of the eyes of the test pair associated with the best score as actual positions of the eyes of the face.
A method for progressively correcting positions of eyes of a face in a two-dimensional digital image on a basis of a pair of approximate positions and an adjacent area around each approximate position, spatial dimensions of the two-dimensional digital image consisting of a first spatial dimension and a second spatial dimension, the first spatial dimension being orthogonal to the second spatial dimension, the method comprising an iteration of the following steps, dimensions of the adjacent area being reduced in each iteration:
- determining test pairs of the positions of the eyes on the basis of the pair of approximate positions, each position of an eye in a test pair being in the adjacent area of the approximate position of a corresponding eye;
- geometrically transforming, for each test pair, the two-dimensional digital image into a normalized image in such a way as to make the positions of the eyes of each test pair coincide with a pair of predetermined reference positions;
- comparing at least a part of each normalized image covering at least a part of the face with a corresponding part of a predetermined image, each comparison providing a score representing a level of resemblance between the part of the normalized image and that of the predetermined image;
- and
- declaring the positions of the eyes of the test pair associated with the best score as new approximate positions of the eyes of the face.
Dependencies are inferred from the claim text.