TV Based Frame Fusion

Forensic imaging practitioners use algorithms and filters borrowed from the traditional "smooth" regularization image processing framework, which is not appropriate to the singular nature of imaging information. A new Total Veriation based, feature-accurate approach to basic image processing tasks has been developed by Cognitech researchers. These image processing tasks include video restoration, frame matching, expansion, sharpening, and image measurements. Numerical instability is overcome by computing solutions of time-dependent, nonlinear PDE's with or without degradation constraints. Moreover, TV based image processing delivers accurate estimation of singularities, i.e. computed image features are a numerically accurate approximation of true image features. Only when the accuracy of all singular image features is assured can an algorithm be used as a serious investigation tool for identification of subjects within the surveillance video. Real life surveillance videos are low resolution, noisy, and blurry. The subject to be identified may be captured by only a small pixel area on each video field. We are proposing an analytic and nonlinear PDE-based fast computational discrete technique for "frame fusion", which is multiscale, accurate, and occlusion capturing. This technique will incorporate video degradations such as motion, out-of-focus atmospheric blurs, as well as recording noise and distortions.