Research On Single Image Super-resolution Reconstruction Using Multi-dictionary Learning

Image is an important carrier of information transmission.However,due to some factors such as equipment and environment,the image is the low-resolution image with noise and lack of details.The cost of obtaining high-resolution image by improving equipment is high.Therefore,the image super-resolution reconstruction technology is the focus of image processing field,and has been widely used in remote sensing,monitoring,medical and other fields.Image super-resolution reconstruction is based on one or more low-resolution images by means of algorithm to estimate a high-resolution image.In order to recover more lost details and improve the effect of super-resolution reconstruction of a single image,this paper proposes a multi-dictionary learning based on support vector regression and an improved iterative back projection algorithm as a post-processing process.The main work and innovations in this paper are as follows:(1)The research background,significance,current development status and theories of various common algorithms of image super-variability reconstruction are introduced in detail,and the advantages and disadvantages of the algorithm are briefly described.(2)Aiming at the problem that learning-based algorithms often use the same eigenvalues in learning dictionaries,a method is proposed that different eigenvalues are used in discrete cosine transform domain and spatial domain,respectively,after raster scanning to learn low/high frequency image patch dictionaries through support vector regression.In predicting phase,to get the regression image,the same method is applied to extract the feature of the test image,the characteristics of different parts of the image is more accurately characterized.(3)In order to reduce the error of the regression image in the reconstruction process,an improved iterative back projection algorithm is proposed as the post-processing process of the image algorithm.Replacing the iterative image in the traditional iterative back projection with the image obtained by support vector regression each time.To avoid the problem that the projection matrix is difficult to determined,the bicubic interpolation is used to replace the iterated back projection matrix.(4)To prove the validity of the proposed algorithm,there are three kinds of experiments: the selection of appropriate dictionary,the comparison of the proposed algorithm with other algorithms,the comparison of improved iterative back projection with traditional iterative back projection.The objective and subjective aspects show that the proposed algorithm improves the image reconstruction effect.