摘要
为了提高单幅图像超分辨方法的性能,提出了新的基于投影矩阵的超分辨率方法。通过引入字典的互不一致性增强求解出的字典的表达能力;对分类后的低分辨率图像特征和相应的高分辨率图像特征的关系学习多个投影矩阵;并通过设置权重矩阵,增强邻近字典原子对当前图像块的表达能力,弱化较远原子的表达能力。在对投影矩阵进行正则化约束的前提下,利用字典原子和特征块之间的相关性以及特征块与其聚类中心的欧式距离关系,重构出拥有更加详细信息的高分辨率图像特征。实验结果显示,该方法的重构结果优于传统方法的重构结果。
In order to boost the performance of the single image super-resolution methods, a new super-resolution method based on projection matrices is proposed. It introduces the item of the mutual incoherence to obtain the atoms which have more representation abilities, and then it learns multiple projection matrices according to the relationship of classified image features between low-resolution images and the corresponding high-resolution images after classification, and a weight matrix is set to increase the representation abilities of the close neighboring atoms and decrease the representation abilities of the atoms which are relatively far away. Under the constraints of the projection matrices with regularization term, the coherence properties between the atoms and feature blocks, and the Euclidean distance relationship of the feature blocks and the classification centers are utilized to reconstruct high-resolution image features with better detailed information. The experimental results show that the reconstruction result of the proposed method is better than that of the traditional method.
作者
端木春江
代晓东
Duanmu Chunjiang;Dai Xiaodong(School of Mathematics and Computer Science,Zhejiang Normal University,Jinhua,Zhejiang 321004,China)
出处
《计算机时代》
2020年第2期1-5,共5页
Computer Era
基金
国家自然科学基金项目(61401399)
浙江省自然科学基金项目(LY15F010007和LY18F010017)
关键词
图像处理
图像超分辨率
字典学习
投影矩阵
互不一致性
image processing
image super-resolution
dictionary learning
projection matrices
mutual incoherence
