From the visibility graph (VG) to a complementary visibility graph (CVG) and an image complementary visibility graph (ICVG): Algorithms and applications

Abstract

A new algorithm for bridging the gap between time series and networks is proposed in this short paper called the complementary visibility graph (CVG). The visibility graphs (VGs) method makes it easy to fulfill complex network topology modeling, which is effective for nonlinear dynamic analysis. Based on the proposed CVG, an image complementary visibility graph (ICVG) is also proposed. The algorithmic procedure has three steps. First, the texture images were converted into the corresponding ICVG. Then, the feature descriptors of the texture image datasets were extracted from the associated complex network set. Finally, texture image classification can be successfully achieved by using the most popular classifiers. Experimentally validated on the classic datasets Kylberg and KTHTIPS2b. The results show that the proposed ICVG model and cubic support vector machine classifier on the two datasets have classification accuracies of 100.0% and 93.0%, respectively. On the same image datasets, the results are better than most results in the existing literature, easily extending to similar situations. The source code is available at https://github.com/LaifanPei/CVG.