一种新型的基于 KCS 算法在图像重构中的应用-《计算机技术与发展》

文章信息/Info

Author(s):: LIU Yan; QIAN Yang; LI Lei; Nanjing Tech University Pujiang Institute,Nanjing 211134,China

Keywords:: kernel dictionary learning; adaptive kernel K-SVD algorithm; kernel compressed sensing; reconstruction time; peak noise ratio

摘要:: 由于原始信号能够通过字典原子的某一线性组合进行稀疏表示,因此在压缩感知理论中,原始的高维信号可以从低维测量值中进行恢复。但是,对于一些信号,譬如图像、视频等,因其具有高维性、多变性以及繁杂性等特点,用线性表示模型难以对其进行稀疏表示。这种情况下,需要在非线性流形下获取更优的稀疏表示。文中首先介绍了核字典学习方法中的 KKSVD 算法,对其稀疏编码阶段进行改进,得到自适应核 K-SVD 字典学习算法(AKKSVD),并将其与核压缩感知理论(KCS)相结合,提出了一种基于 AKKSVD 字典学习的 KCS 算法(AKKSVD-KCS)。通过对图像进行重构的仿真对比实验表明,该算法对非线性信号的重构更具备高效性,相较于其他算法在重构时间、峰值信噪比等方面更具有优越性,即其重构性能更佳。

Abstract:: Since the original signal can be represented sparsely by a linear combination of dictionary atoms,in the compressed sensing theory,the original high-dimensional signal can be recovered from the low-dimensional measured value. However,for some signals,such as images and video,it is difficult to conduct sparse representation with linear representation model due to their characteristics of high dimensional,variability and complexity. In this case,it is necessary to obtain better sparse representation under nonlinear manifolds. First the KKSVD algorithm is introduced in the kernel dictionary learning method,and its sparse coding stage is improved to obtain the adaptive kernel K-SVD dictionary learning algorithm (AKKSVD). Combining it with the nuclear compression perception theory (KCS),a KCS algorithm (AKKSVD-KCS) based on adaptive KKSVD dictionary learning is proposed. Simulation and comparison experiments on image reconstruction show that the algorithm proposed is more efficient in the reconstruction of nonlinear signals,and has more advantages in the reconstruction time and peak noise ratio compared with other algorithms,that is,it has better reconstruction performance.