Abstract: Recently, double dictionary training has emerged as a powerful tool for solving a class of super-resolution reconstruction problems in computer vision and image processing. To reduce the reconstruction error of dictionary training and reconstruction within image super-resolution reconstruction algorithm via double dictionary training, this paper proposes a super-resolution reconstruction algorithm based on adaptive multiple dictionary pairs (AMDP). This algorithm is available for different types of input images by samples clustering and training features dictionary. In dictionary training, differences among different dictionaries training are used to filter high frequency patches and reconstruct images repetitiously in restoration, which effectively improves the quality of reconstructed images. The experiments show that the proposed algorithm achieves improvement in image quality and provides more details.

Key words: sparse representation, super-resolution, sample cluster, dictionary pair, dictionary training

摘要： 最近，双字典训练已成为在计算机视觉和图像领域解决超分辨率复原问题的有力工具。针对基于双字典训练的图像超分辨率算法中字典训练与重构阶段的重构误差，提出了一种基于自适应多字典对的超分辨率复原算法。通过对样本进行聚类并训练多特征字典来适应不同类型的输入图像。在字典训练阶段，充分利用了不同次训练字典产生的差异，在重建中筛选高频补丁，进行多次重构，有效地提升了重构图像的质量。实验仿真与比较表明，该方法在重构图像的质量上有所提高，且能提供更清晰的细节。

关键词: 稀疏表示, 超分辨率, 样本聚类, 字典对, 字典训练