基于聚类和回归的超分辨率方法

Super-resolution via Clustering and Regression

基于实例的超分辨算法在超分辨重建技术中得到了广泛的应用，如邻域嵌入算法。这个算法的理论基础建立在流形假设上．而流形假设往往并不是成立。为了避免流形假设，提出了基于聚类和回归的超分辨率方法。由于图像的信息较为复杂，因此首先将图像样本块进行聚类。然后，在高分辨图像块与低分辨图像块中学习局部回归函数。最后应用回归函数进行超分辨重建。将仿真实验结果与其他算法进行比较，证明了该算法的可行性和有效性。

Example-based super-resolution algorithm has been widely used in the super-resolution reconstruction technology, such as neighbor embedding algorithm. These algorithms are based on manifold assumption, while manifold assumption does not always hold. In order to avoid the assumption, a super-resolution method based on cluster and regression is presented. Because of the complexity of image information, firstly, the image patches are clustered. Then, a local function is learned between high-resolution image patches and low-resolution image patches. Finally, the function is used to generate high image. The result of simulation experiment, which is compared to other algorithms, indicated the feasibility and the effectiveness of this algorithm.