Graph reduction method using localization operator and its application to pyramid transform
Graph reduction method using localization operator and its application to pyramid transform
复制标题
基于定位算子的图约简方法及其在金字塔变换中的应用
DOI：
10.1109/apsipa.2017.8282001
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
Y. Tanaka
中科院分区：
文献类型：
--
作者：
A. Sakiyama;Y. Tanaka
针对图上定义的信号，提出了一种利用局部化算子的图约简方法。在以前的工作中，我们提出了一种方法，传感器的位置选择考虑信号的局部化在图的顶点和频谱域。本文将选择方法推广到图的约简。它根据图的截止频率选择下采样的节点，使得下采样的信号保留了大量的信息，用于预测去除节点上的值。此外，它还考虑了在图的顶点域的本地化。利用图约简思想设计了多尺度分析方法，并将其应用于图信号的去噪。
This paper proposes a graph reduction method by using the localization operator for signals defined on graphs. In the previous work, we proposed a method for sensor position selection considering signal localizations in graph vertex and spectral domains. This paper extends the selection method to graph reduction. It selects downsampled nodes according to the graph cut-off frequencies so that the downsampled signal preserves the much information for predicting the values on removed nodes. Furthermore, it also considers the localization in the graph vertex domain. We design the multiscale analysis with the graph reduction and apply it to graph signal denoising.