This proposal responds to DARPAs STTR solicitation, topic 7 (ST082-007). Advanced computational technology based on solid mathematical ground will be developed and applied to elimination of blurs and super-resolution of video of faces. These applications are important in aerial surveillance systems and homeland security tasks. Success of the project will overcome limitations of current methods in terms of their accuracy and efficiency. The method is significantly more advanced than the existing techniques mainly because it is based on a new motion estimate approach, which determines pixel correspondence between two video frames by optimizing a similarity measure subject to constraints expressed in terms of differential operators. The imposed constraints guarantee that the admissible space consists of all smooth and invertible transformations. Thus, the approach is capable of detecting nonlinear, complicated motions that are typical in a video sequence.
Keywords: Motion Blurs, Super-Resolution, Video Of Faces, Multigrid Optimization
