Pocs-based Image Super-resolution Reconstruction Algorithm

Super-resolution image reconstruction technology belongs to the area of the image restoration. The traditional method of image restoration processes only one low-resolution frame which contains less information, and the quality of the restored image is also low. Thus this method is inefficient. However, Super-resolution image reconstruction technology using multiple frames which contain relative motion, blur, under sampling and noise pollution to reconstruct one high-resolution image. It is possible that the reconstructed image has higher resolution than the original image because the multiple frames contain different information of the details.At present, the researchers are seeking and exploring the math model, theoretical framework and efficient calculation of the super-resolution image reconstruction. While trying to comprehensively analyse the super-resolution image reconstruction technology, the paper especially explored and studied super-resolution reconstruction method based on POCS, improved the method of selecting the reference frame and the algorithm of motion estimation and finally presented the view of the research and experiment results. The whole work of the paper is summarized as follow:Ⅰ. Compared and studied the domain and space domain algorithm of super-resolution reconstruction method. Meanwhile compared and studied POCS and MAP algorithm of the space domain algorithm. The result shows that POCS is stronger to maintain the edgy, faster to constringency and more convenient to compute. Thus the paper did further study in POCS and mainly in two important parts of the POCS:motion estimation and construction method of the reference frame.Ⅱ. The paper proposed an improved square-diamond-searching motion estimation algorithm which not only maintain the Advantages of the block matching algorithm, but also improved square-diamond-searching algorithm on the basis of the fact that the best match is surrounding the center point due to the small movement of the relative frames. This method using large square template to set four vertices, and then using small diamond template to search around the matching point after the matching points being determined. The experiment results show that on premise of ensures the accuracy of the motion estimation the improved searching algorithm is more or less faster than TSS, NTSS, FSS, DS and SDS.Ⅲ.In response to this shortcoming that POCS is highly dependent on the initial estimate, the Paper proposed an interpolation method based on Fractal and wavelet transform to interpolate the reference frame. Compare with bilinear interpolation, this method reduce the blur of the detail. Thus the reconstruction result of POCS using this method is better than using bilinear interpolation. Finally, the paper proved the algorithms'feasibility and validity by experiments.