| 摘要 |
<P>PROBLEM TO BE SOLVED: To estimate a shape of a face, on which a three-dimensional expression change difficult to detect with only two-dimensional images has occurred, with high accuracy and low computational complexity. <P>SOLUTION: A face texture evaluation value calculating unit 103 calculates a face texture evaluation value representing a degree of matching between an input face image and a face shape candidate texture for each of a plurality of face shape candidates. A face expression change likelihood calculation unit 104 calculates a first likelihood between a reference face shape and the face shape candidate for each of the plurality of face shape candidate. A correlation evaluation unit 105 calculates a first correlation evaluation value representing strength of correlation between the face texture evaluation value and the first likelihood for each of the plurality of face shape candidate. An extraction unit 107 extracts a face shape candidate as a measurement object for which the first correlation evaluation value is lower than a first threshold, out of the plurality of face shape candidates. <P>COPYRIGHT: (C)2013,JPO&INPIT |
