Video enhancement using an iterative multiframe SRR based on a robust stochastic estimation with an improved observation model

Abstract

This paper proposes a video enhancement method using a novel super-resolution reconstruction (SRR) framework for real standard sequences that are corrupted by any noise models. The traditional SRR algorithms are very sensitive to their assumed model of data and noise, which limits their utility. The real noise models that corrupt the measure sequence are unknown; consequently, SRR algorithm using LI or L2 norm may degrade the image sequence rather than enhance it. The robust norm applicable to several noise and data models is desired in SRR algorithms. First, this paper proposes a robust SRR algorithm based on the stochastic regularization technique of Bayesian MAP estimation by minimizing a cost function. The Huber norm with Tikhonov regularization is used for measuring the difference between the projected estimate of the high-resolution image and each low resolution image, removing outliers in the data. Second, in order to cope with real sequences and complex motion sequences, this paper proposes an improved SRR observation model, affine block-based transform, devoted to the case of nonisometric inter-frame motion. The experimental results show that the proposed reconstruction can enhance real complex motion sequences, such as Suzie and Foreman sequence, and confirm the effectiveness of our algorithm and demonstrate its superiority to other SRR algorithms based on LI and L2 norm for several noise models such as AWGN, Poisson, Salt&Pepper and Speckle noise.