Image Super-resolution Reconstruction Based On Nonlocal Self-similarity

With the further improvement of m odern science and technology information level,we have finally ushered in a new information time.As a simple and direct way of informalion acquisition,image has a very important application in many fields.Among all kinds of information in the image,the resolution represents the amount of information stored in the image.The higher the resolution is,the more detailed the image contains.However,in real life,under the influence of imaging equipment and environmental factors,high-resolution images are often not available,and hardware technology often has high cost and great technical difficulties,so image super-resolution reconstruction technology as a software technology which can effectively improve image resolution came into being.At present,although the image super-resolution reconstruction algorithm has achieved good research results,the image edge maintenance and the robustn ess to noisy image are still the key issues of reconstruction technology.Therefore,this paper will use the important image prior knowledge of image nonlocal self-similarity as the research main line,the following results are obtained from the research of image edge directionality and similar block matching(1)In order to preserve the directional feature of the edge and suppress the edge artifacts,considering the multi-directional feature information of the image,this paper proposes a reconstruction algorithm based on the directional feature of the image.The algorithm is a super-resolution reconstruction framework based on the regular terms of total variation and non-local means.However,the total variation algorithm can not keep the directional features of the edge well,and the non-local means method does not consider the local geometry of the image block.The proposed algorithm in this paper firstly uses the directionlet which can decompose the image in multiple directions to extract the directional feature image,and then according to the DTV which can better retain the characteristics of th e edge in the vertical and horizontal directions,rotates the extracted directional feature to the horizontal and vertical direction to construct the total variation of the modified direction to maintain the edge direction feature.Next,the direction feature of each pixel is added to the similar weight estimation in the nonlocal means method,which can effectively improve the matching accuracy of similar blocks.The algorithm adds the direction feature of the image to the regular constraint,which can effectively preserve the information related to the geometry of the image and also suppress the edge artifacts.(2)In order to improve the robustness of the reconstruction algorithm based on gradient profile to the noisy image,BM3D algorithm is used to filter the low-resolution noise image,and then the reconstruction algorithm is used.Based on the fact that the BM3D algorithm is not flexible enough to search the region in the block matching stage and can not make good use of the similarity of the local structure of the image,an improved method is proposed to search the super pixel region with high similarity of the pixels in the region.Firstly,SLIC algorithm is used to segment the image.Then,in the block matching stage of BM3D algorithm,super pixel is used as the search box to search,and the resulting 3D matrix is used to complete the filtering operation of noisy image.Next,the denoised image is used as the input image of reconstruction,and the gradient profile based algorithm which can enhance the edge detail information is used for reconstruction.The evaluation method of subjective vision and objective analysis proves that the algorithm can improve the reconstruction effect of noisy image and keep the edge of image well.