| 摘要 |
PROBLEM TO BE SOLVED: To reduce burdens on a user without contacting, to be solid relating to the rotation of a face and to mitigate the limitation of an operation. SOLUTION: The face image of a testee is picked up by a CCD camera and the CCD camera is controlled so as to match the centroid of a skin color area and the center of an input image in an image pickup control part 7. Also, in a feature point detection part 4, the positions of the eyes and the mouth are estimated from face image signals. The inclination angle of the face is estimated based on the estimated positions of both eyes, the face image inside a detection area is rotated for the angle and an expression estimation part 5 performs two-dimensional discrete cosine transformation to the rotated detection area and obtains the change of spatial frequency components from the time of non expression detection. In an expression reproducing part 6, by using a parameter learned by genetic algorithm beforehand, the obtained change data of the spatial frequency components are transformed to the deformation of a three-dimensional face model and an expression is reproduced.
|
