Optimization of the texture‐geometric image model for estimation of the face parameters

Consideration was given to an approach to describe the image texture and geometry by expansion in the parameterized system of basic functions. Parametrization allows one to construct the basis in some designated model coordinate system. The best basic space is found by optimizing the parameters of basic functions in terms of the learning set of images of the object under consideration. After the learning stage, each texture of the learning set is described by weight vector, set of the characteristic point coordinates, and assembly of the qualitative characteristics defined by an expert in the model coordinate system. The geometry of the image under consideration is estimated by minimizing the residual of the model weight vectors and projections of the deformed image in the basic space, and at the same time the closest texture pattern is determined. This approach is used to construct the texture‐geometric model of personal images, as well as a composition of models of images of the face elements. The models obtained are used to solve the problems of face tracking and localization the face element from the video images.