Research And Improvement On The Person's Face Image Deformation

As one of the popular information media, image is the carrier, with which people can sense the world and express various messages. With the rapid development of computer technology, more and more researches on image processing are conducted. Based on certain transformation functions, smooth and nature image can be achieved with controling the feature points. Methods of face image transformation have are researched. Detail research contents and experiment results are as follows:1. An algorithm for square grids image morphing is proposed. With this algorithm, square grids are created in face image and covered the whole image. When one or more square vertex is changed, the others' shapes are also changed because of the mapping transformation relationship between them. After changing control point and establishing corresponding relationship of the 4 groups'points from 2D to 2D, the transformation matrix can be achieved by DLT algorithm and a controllable face image morphing can be realized.2. Different from traditional methods, an algorithm based on triangulation is improved to realize gradual cross change between two face images and proved to be with more controllability and selectivity. By choosing different face feature points and certain amount of control points, the whole transformation process can be controlled. Feature points are selected arbitrarily in the feature region of Delaunay which is defined with triangle grid. For the problem that the calculation of triangulation is time-consuming, an improved algorithm based on Bresenham coordinate transformation is applied when calculating the pixel coordinate in the original image which is corresponding to the pixel in the current triangle. Operations of addition and subtraction are adopted in this improved algorithm of Bresenham, and the multiplication and round operation can be avoided. Therefore, image morphing can be speeded up.3. In reference to the standard of MPEG-4,68 face feature points are extracted by using feature point extracting system based on ASM. Then, based on MLS theories, the mathematics model of image morphing is analyzed. Affine transformation function based on point set, similar transformation function and rigid transformation function are deduced respectively. Rigid transformation algorithm is combined with facial expression rules in the experiment. According to the distribution of 68 feature points in face, the same amount feature points with different position are selected in twice to generate morphing image of the target. The experiment proves that with this method, not only the visual effect is more fluent, but also the sense of reality is enhanced.