动态网络分析视角下的纹理分析与分类应用研究

Texture Analysis and Classification Application Under the Viewpoint of Dynamic Network Analysis Strategy

基于Backes Andre Ricardo等人提出的“构造伴生动态演化网络”、生成“高维解释向量”的纹理图片分类法,研究了Kylberg纹理图像库的分类问题。通过将数据集图片样本进行二次分割以减少该算法预处理计算量、提出了构建纹理网络时的“网络顶点随机化、抽样采集”策略,该思路可进一步减少算法的时间开销。该类纹理图片数据高维几何数字特征的提取流程易于实现,扩展了原算法的适用范围。数值结果表明算法对旋转操作、噪声干扰具有一定的鲁棒性,具备一定潜在应用前景。

Based on the newly proposed texture analysis method on shape analysis via constructing explanatory variables with associated dynamic evolutionary complex networks proposed by Backes Andre Ricardo et al,this paper studies the classification problem of Kylberg texture image library.The image samples of the dataset are segmented twice to reduce the computational complexity of the algorithm.An improved strategy of"network vertex randomization and sampling collection"is also proposed for further reducing the calculation time on CPU.The extraction process of high-dimensional geometric digital features of this kind of texture image data is easy to be implemented,which extends the scope of application of the original algorithm.Numerical results show that the algorithm is robust to rotation operation and noise interference.It has potential application prospect in related research fields.