Video Processing and Regularized Inversion Methods
Guy Le Besnerais, Frédéric Champagnat · 2015
This chapter demonstrates the significance of methodologies derived from regularized inversion for video processing and more generally for low level vision, to the extent that a preliminary reflection is conducted on an adequate observation model. It illustrates three applications: motion estimation from images, multiview stereovision, and superresolution (SR). Particle Image Velocimetry (PIV) is an experimental technique of fluid mechanics that enables the measurement of velocity fields. Stereovision is a photogrammetric technique for the estimation of geometrical quantities relating to a 3D scene from two views taken from two different positions. SR consists of using the temporal redundancy of a video to estimate an image with better quality and, in particular, better resolved. In all three cases, a common problem of image registration is revealed. The chapter shows achievements that are based on choices and approximations of the data fitting term to obtain effective techniques in a given application context.