Skip to Main Content
This paper presents the application of a robust extended Kalman filter (REKF) for an image processing problem in order to mitigate the effects of noise introduced during registration of geometrically warped anisoplanatic images. Emphasis is laid upon maintaining accuracy at the pixel-level. The perceived wander (random motion) of image intensities at the pixel level is modeled as a second-order oscillator, with a time-varying uncertain natural frequency. A robust extended Kalman filter (REKF) is used to estimate the state of the uncertain system and restore the pixel wander from a set of synthetically generated noisy measurements of registration shiftmaps. Also, a comparison of REKF and standard EKF estimation methods is presented.