A Super-resolution Algorithm through Neighbor Embedding

Super-resolution algorithm through neighbor embedding (NE) is based on learning. But the algorithm have the problems that the feature of image is complex and difficult to classify and search. A super-resolution algorithm through neighbor embedding by the feature based on the ratio of multiple directions of two-step grade is proposed. The computational complexity of the feature is low, and classifying and searching the feature is simple. Meanwhile the memory space of the algorithm is small. The super resolution algorithm is easy to implement on hardware. The experiment results shows that, high resolution images reconstructed by the algorithm have a great improvement than other state-of-the-art SR algorithms, also subjectively recover more high frequency detail and sharp edge. The recover images have better subjective and objective quality.