Research On Image Super-resolution Reconstruction Algorithm Based On Sparse Representation

Image super-resolution reconstruction technology can generate clear high-resolution image from low-resolution image and is widely used in many areas of real life.The sparse representation based image super-resolution algorithm has attracted many researchers' attentions because it has better reconstruction performance and is more robust than the traditional super-resolution algorithms.The classic sparse representation based algorithm is too time-consuming,may result in block artifacts near the edges and cannot reconstruct clear texture details of intricate natural images.To deal with these defects,this paper presented two improved methods:Firstly,due to the textural similarity between high-resolution image and its lowresolution version,an image super-resolution reconstruction method based on sparse representation and local texture constraint is proposed.The local texture consistency constraint between high-resolution and low-resolution patches is added to local patch sparse representation model to improve reconstruction fidelity.In the image postprocessing process,the global reconstruction optimization model and the non-local selfsimilarity optimization model are used to further optimize the reconstruction results.Experimental results on noiseless and noisy images show that the proposed method can recover more texture details and has good noise robustness.Secondly,an image super-resolution algorithm via sparse representation of LBP feature is proposed.Local binary patterns are used to extract the texture features,the training patches is classified by K-Means to learn the over-complete dictionary for each class.In local patch reconstruction model,we use the sparse representation of the LBP feature of low-resolution patch to reconstruct high image patch.In the global model,a consistency constraint of the normalized LBP histogram feature between high-and lowresolution images is added to enhance the result.Experimental results demonstrate that the proposed method cost less time to reconstruct high-resolution image and has better performance in recovering texture details than and the first method.