Research On Neighbor Embedding Based Image Super-Resolution Reconstruction

Super-resolution (SR) is originated in the area of image restoration. Increasing the image resolution by hardware is more and more difficult, it is more widely to use SR reconstruction technology to enlarge the image and increase details. SR reconstruction is a software technology based on signal processing. Because it has nothing to do with the hardware and low cost. It has a broad application prospects in the fields of remote sensing, video, medical, public security, network and so on.This article firstly describes the research background and current situation of the SR reconstruction. Then, introduce the basic principle of the SR reconstruction, which can indicate the feasibility of SR reconstruction. Three categories of SR reconstruction algorithms are also introduced in detail. This article focuses on the neighbor embedding SR method. The major work is as follows:(1) A natural image always contains many similar image blocks. The redundant information can be used to solve various problems in image processing. Non-local similarity is based on this feature. This article leads to the basic concept of neighbor embedding by introducing the manifold learning. Then, based on the original neighbor embedding SR method, we propose a new feature, and join the non-local similarity to the algorithm. Based on these improvements, a novel neighbor embedding SR method based on non-local similarity constraint is proposed.(2) The SR reconstruction technology can be applied to the network transmission. This application has two features:compressed images and real-time requirement. This article uses k-means algorithm to classify the image training set, which can reduce the complexity of the search for matching and improve the algorithm speed; we also introduce the new features and create a new solving formula for reconstruction coefficients, which can enhance the reconstruction quality. At last, the image reconstruction demo interface based on windows is presented, indicating the algorithm’s ability for practical application.Finally, combine and conclude the content of this article. The future research direction is also proposed.