Research On Super Resolution Reconstruction Of Image Based On Sparse Representation And Regression

It is difficult to obtain an ideal high resolution image because of the limitation of the technical cost of the imaging equipment and the external environment.The method of image super resolution reconstruction is to use the digital signal processing technology to reconstruct higher resolution images with the lower resolution images without changing the conditions of the current equipment technology and cost.The method of super resolution reconstruction based on sparse representation of the learning methods is the most hot research direction recently which can obtain the higher quality images than the traditional methods.It's one of the most effective reconstruction methods and has been used in many fields such as video surveillance,medical care and so on.The image super resolution reconstruction was studied by the reconstruction method based on the sparse representation in this paper.According to the sparse representation coefficients of the high and the low resolution dictionaries used in the traditional reconstruction methods based on the sparse representation are the same without considering the nonlinear differences between the high and the low sparse representation coefficients,making the result of image resolution reconstruction lose some high frequency information.And the K-means clustering method used to cluster the images generally is too dependent on the initial value in the process of reconstruction,which leads to the clustering results is not stable within a certain range and reduces the robustness of reconstruction results.So a reconstruction method based on the sparse representation and support vector regression was proposed under the premise of adaptive initial center points classification in this paper.The images were clustered by the clustering method of adaptive initial center points,the obtained clustering results were fewer iterations and more stable.Then,the sparse representation coefficients of various types of the high and the low resolution dictionaries were used as the samples and got the training models of the relationship between all kinds of the high and the low sparse representation coefficients trained by the support vector regression.The sparse representation coefficients of the low resolution images to be reconstructed were used to predict the high sparse representation coefficients of the corresponding high resolution images by the training models of the corresponding classes in the reconstruction phase,recombined with the corresponding high resolution dictionaries reconstructed the high resolution images.It is better to restore the missing high frequency detailed information compared with the previous reconstruction methods.The higher the quality of the reconstructed images,the better the robustness.There is a certain lack for the image super resolution reconstruction by using a single dictionary to restore the high frequency components,an improved reconstruction method of sparse representation and regression based on double dictionaries which based on the reconstruction method of the sparse representation and support vector regression was proposed in order to dig out the missed high frequency parts of the low resolution images as much as possible.The high frequency information was divided into the main high frequency components and the secondary high frequency components in the process of reconstruction,then the main dictionaries and the secondary dictionaries were obtained by training the sample set in the two stages of corresponding to the main high frequency and the secondary frequency in the training process,collectively referred to as double dictionaries.The low resolution images were reconstructed step by step through using the double dictionaries.From the experimental results,whether subjective visual experience or objective quantitative analysis of the data,the reconstruction method of sparse representation and regression based on the double dictionaries can recover more high frequency detailed information compared with other reconstruction methods.The better the quality of the reconstructed image,the higher the clarity.