A Leclerc Bayesian approach for video reconstruction based on a robust iterative SRR and a General Observation Model

Abstract

Traditional Super-Resolution Reconstruction (SRR) vigorously falls back on the availability of accurate registration for this fusion task and the observation noise model. When the motion is registered inaccurately, as often happens for nonglobal motion fields, annoying artifacts appear in the super-resolved outcome and when the observation noise is not AWGN, severe artifacts appear in the reconstructed result. This paper proposes the alternative robust SRR algorithm that can be successively applied on the real or standard sequence and can be applied on the sequences that are corrupted by various noise models. First, the proposed SRR algorithm is based on Bayesian framework with the Leclerc norm for measuring the error between the projected estimate of the high quality reconstructed image and each corrupted images and for removing outliers in the data. Second, the proposed algorithm is used a General Observation Model or GOM (or fast affine block-based transform) in order to cope with real complex motion or nonisometric inter-frame motion sequences. The experimental results demonstrate that the proposed algorithm can be well applied on real sequences such as Suzie and Foreman sequence at several noise models (such as AWGN, Poisson, Salt & Pepper noise and Speckle) and several noise power.