Based On The Gradient Information Of Image Sampling And Super-resolution Reconstruction

In the past dacade, the Super Resolution (SR) reconstruction technology has been playing an important role in image processing field since it was introduced, and thousands of SR papers have bloomed into publications. With the development of the research of computer vision and pattern recognition, there are more and more needs of high resolution for real applications, resulting in super resolution which is based on digital signal processing being focused by many researchers and scientists. In order to adapt to different kinds of applications, super resolution technology should be compact and efficient which means it should have a low computation cost meanwhile keep excellent reconstruction result.Image super resolution via sparse representation not only has good reconstruction result but also has a compact representation. However, it still has some shortness to be solved. Hence, we developed an algorithm based on super resolution via sparse representation.Our work mainly analyzed the merits and shortness of super resolution via sparse representation. We introduced the theory of sparse representation mathematically and the overcomplete dictionary design. Besides, we applied sparse respresentation in super resolution and compared this method with other classic super resolution algorithms to demonstrate the reconstruction ability of super resolution via sparse representation. Due to the highly construction of sparse representation and the fact that the algorithm based on image patches, the results of super resolution via image representation have jiggy and jaggy arifacts in the edges of image. Hence, we proposed an algorithm which induced the gradient imformation to enhance edge and eliminate the artifacts in the edges. We proposed a1D gradient profile as a prior knowledge to represent the edge imformation in image and model the gradient profile. In order to make gradient profile applied in super resolution, we analyzed the relation of gradient profile prior between high resolution image and low resolution image. According to this statistic relation, we proposed a super resolution algorithm based on gradient imformation to enhance the edge in reconstruction image. At final, we designed experiments to test our algorithm. It is pledged by the experiment results that the proposed method is truly effective.