Research On Super-resolution Reconstruction Algorithm Of Single Frame Image

70%-80% of information acquired by humans comes from vision.Images play an indispensable role in information transmission.High-quality and high-resolution images mean more abundant of image details.In the early days,The Super-resolution Reconstruction algorithm was proposed in order to get rid of the high-resolution hardware and improve the image resolution through software.Nowadays,although the hardware of image has been greatly improved,people's demand for highresolution images is also rising.Super-resolution reconstruction algorithm still has application market and research significance.With the increase of computing power,the Single-Image Super-resolution Reconstruction algorithm based on sample learning catch many researchers' interest.Compared with the classical super-resolution reconstruction algorithm,the former one can reconstruct the image with a more complex image structure.However,most SISR algorithms use the external sample database will raise a potential problems that the mismatch of image feature.In contrast,the super-resolution reconstruction algorithm using the internal sample database often only has a small sample to learn the priori information.To solve this problem,we do some innovative research word in the third chapter.(1)the geometric deformation of the image patch is used to extend the size of the database,so the database can more fully express the transformation of the scene and texture appearance;the new formula that measure the similarity of image patch is designed to add the scale of the image patch,direction and brightness as the similarity measure,that improve the accuracy of the similar image patch search.In addition,In the algorithm of search the similar image patch,The algorithm of PatchMatch is optimized,and add the Simulate Anneal algorithm to increase its ability to jump out of local minimum.In addition,the super-resolution algorithms tend to use larger database and more complex algorithms to improve the quality of reconstruction,the running time of the reconstruction algorithm ofen be ignored.In this paper,the fourth chapter makes the following innovation research:(2)the neighborhood embedding algorithm and sparse representation are combined to study the dictionary,to improve the performance of the feature dictionary and to obtain more features.a large number of computing processes are placed in the dictionary training stage,and the algorithm tests only need to search adjacent anchors,and then map to high resolution space;geometric transformation of the input low resolution images,We use small multiple number to iterate to generate the required magnification number.The experimental results show the image reconstruction quality of two superresolution reconstruction algorithms that the third and fourth chapters of this paper proposed is better than most classical method,both in subjective and objective indexes;the fourth chapter has a significant improvement in running time and is feasible.