Automatic Generation of Facial Expression Using Triangular Geometric Deformation

This paper presents an image deformation algorithm and constructs an automatic facial expression generation system to generate new facial expressions in neutral state. After the users input the face image in a neutral state into the system, the system separates the possible facial areas and the image background by skin color segmentation. It then uses the morphological operation to remove noise and to capture the organs of facial expression, such as the eyes, mouth, eyebrow, and nose. The feature control points are labeled according to the feature points (FPs) defined by MPEG-4. After the designation of the deformation expression, the system also increases the image correction points based on the obtained FP coordinates. The FPs are utilized as image deformation units by triangular segmentation. The triangle is split into two vectors. The triangle points are regarded as linear combinations of two vectors, and the coefficients of the linear combinations correspond to the triangular vectors of the original image. Next, the corresponding coordinates are obtained to complete the image correction by image interpolation technology to generate the new expression. As for the proposed deformation algorithm, 10 additional correction points are generated in the positions corresponding to the FPs obtained according to MPEG-4. Obtaining the correction points within a very short operation time is easy. Using a particular triangulation for deformation can extend the material area without narrowing the unwanted material area, thus saving the filling material operation in some areas.