Research On The Key Techniques Of POCS Based Super-resolution Video Restoration

Super-solution(SR) image reconstruction is one of the most spotlighted research areas, because it can overcome the inherent resolution limitation of an imaging system and improve the performance of most digital image processing applications. By utilizing estimation theory, SR is employed to enhance the quality of images. It not only possesses significance pertaining to the theoretical study of signal and image processing as well as pattern recognition, but also boasts a wide range of applications including video display for entertainment, medical imaging for diagnostics, remote sensing, military surveillance, video frame freezing, video Transmission etc. This thesis focuses on the study of two main subjects– motion estimation and image restoration– in SR technique. We have proposed an improved SR method based on POCS considering the restriction of motion estimation error and edge information preservation.To begin with, the background, concept, and fundamental problem of super-resolution image restoration was introduced. After a detailed description of degradation incurred in the imaging and compression process through signal and system modeling, the restoration method based on POCS was introduced. Next, in the study and realization of the two motion estimation methods - sub-pixel hierarchy block matching and optical flow estimation based on wavelet projection - the Non-Gaussian distribution character of the motion compensated error was analyzed. We proposed a ratio-comparing method which could not only mitigate the error caused by inaccurate motion estimation, but also avoid discarding the useful information in the projected residual. Following this, we studied the anisotropic diffusion process of imaging for POCS based image restoration. Two methods for estimation anisotropic PSF were proposed. One is based on edge pattern classification, and the other is taking advantage of singularity value decomposition. Both are applicable to steer the PSF, which is the base for image restoration. To make the prior two sets of information self-contained, the convexity of restriction of motion estimation residue and edges are proved. And as a refinement, these restrictions are integrated into the traditional Super resolution restoration procedure, which is based on POCS. The experimental results show that the performance of this algorithm is better than the traditional method, both in visual effect and peak signal to noise ratio.Finally, considering the effectiveness of character representation in Transformation domain in the current study trend, we did further research for SR technique in Transformation domain, and realized two methods. One uses DCT for compressed video restoration and the other is based on wavelet coefficient estimation. The result of the final experiment indicates that our method is applicable to restore the detail information of the degraded image.