Research On Image Super - Resolution Algorithm Based On RLLE

Nowadays, Image super-resolution technology is a very active technology; we can use it to estimate an image that is of a better overall visual effect and a higher resolution from one or more of low resolution images. The technology is widely used in video security monitoring, meteorological remote sensing, medical imaging, etc. Learning-based method is one of important content. We can learn a joint system model by training two training set which include a group low-resolution and the corresponding high-resolution image at the same time. To explore the learning-based method is our direction which uses the manifold theory. We use manifold learning algorithm of nonlinear dimension reduction in this paper. We regard the consistent characteristics of local geometric structure of manifold on the low and high dimensional space as our system model’s basis.Our paper mainly undertakes two projects those are the manifold learning and learning-based super-resolution algorithm. In order to overcome shortcoming of the LLE (NE) algorithm among manifold learning algorithm--the sample is very sensitive to outliers (noise), many scholars put forward different robust local linear embedding (RLLE) algorithm. And with analyzing the classic article that is named "Super-Resolution Thorough Neighborhood Embedding (SR-NE)" presented by Chang and others published in CVPR. Our paper puts forward the super-resolution reconstruction algorithm based on robust local linear embedding(SR-RLLE) which not only has good robustness to outliers, but also get better super-resolution reconstruction results.Our paper consists of the following components.Firstly, our paper summarizes related literatures about LLE algorithm, and analyzes the advantages and disadvantages of LLE and other improved algorithms.Secondly, our paper descript respectively each RLLE algorithm in detail, then summarize and deeply analysis the advantages and disadvantages of each algorithm.Thirdly, we improved perfectly SR-NE algorithm by Chang-RLLE algorithm which is of higher feasibility. In order to overcome SR-NE algorithm shortcoming--sensitive to noise, we propose the SR-RLLE algorithm which effectively solved this disadvantage.Finally, according to the experimental results, analysis and compare to the experimental result of SR-NE algorithm, the SR-RLLE algorithm which we proposed in this paper not only has good robustness to outliers, and but also get better super-resolution reconstruction results.