Study Of Super-Resolution Restoration For Image Sequences

Super-resolution restoration (SRR) refers to restoring a high-resolution and high-quality image from multiple low-resolution degraded observations by signal processing, which can remedy defects of resolution and quality limit caused by the physical and cost restrict of hardware and various natural distortion in imaging process. In recent years, SRR is a hot research issue which can be widely used in pattern recognition, video processing, remote sensing, medical imaging, etc.SRR approach is originally proposed in spectrum domain, however, as the observation model is restricted to global translational motion and LSI (linear shift invariant) blur, the spectrum domain SRR approach is limited extremely, while the spatial domain related approach is much more active research issue in the literature. MAP (maximum a posteriori ) SRR is one of the most popular and effective spatial SRR approach, which is flexible and robust in modeling noise characteristics and a spatial priori knowledge with a existent and unique global optimal solution. In this paper, we mainy discuss MAP SRR issue.In this paper, an edge-preserving adaptive image prior with strong edge-preserving ability and fast convergence speed is proposed for the prior term of MAP SRR frame. The scale and type of the poposed prior can be adjusted manually under various application environment. A SSR algorithm for a single image, a selective image enhancement algorithm, a conventional image restoration algorithm and a Gaussian filtering algorithm are derived from the prior model and MAP frame, respectively.To meet the weakness of conventional MAP algorithms for image sequences, a spatio-temporal adaptive SRR algorithm based on MAP frame is proposed. The spatio-temporal adaptive mechanism, which is induced to MAP SRR frame, can not only preserve edges but also prevent reconstructed image from the influence of inaccurate motion vectors to some extent. Experimental results demonstrate that the proposed algorithm can preserve edges of the reconstructed image effectively with good reconstructed quality and fast convergence speed. Besides, a novel Frame-Iterative SSR algorithm frame for image sequences is proposed in this paper.The sample-blur kernel is proposed to model subsampling and blur jointly related in degrade imaging process, which is used for SRR algorithms applicable for observation model consist of matrixes of subsampling, blur, warping and noising. In order to perform the proposed SRR algorithms on actual blur images, a novel circularly symmetric PSF (point-spread function) recovery method based on cubic spline interpolation is proposed. Experimental results demonstrate that the proposed method can recover circularly symmetric PSF of actual blur images effectively.