【作者】 张磊； 王备战；

【Author】 Lei Zhang, Beizhan Wang ( School of Software, Xiamen University, Fujian 361005, China)

【机构】 厦门大学软件学院；

【摘要】 在充分了解复杂网络基本性质和概念的基础上,先将经过取样和量化得到的数字图像转化为复杂网络的描述形式。得到了数字图像的复杂网络表示,我们就可以设计出更复杂的算法对图像进行特征分析。图象的特征分析不仅依赖于局部特征(如像素和各自邻近像素),并且也有赖于其特征向量。例如,节点度数以及其它有关节点的测量方法(如度或簇系数),可以作为用来描绘纹理或区分空间关系的辅助手段。更多还原

【Abstract】 On the basis of fully comprehending the basic properties and conceptions of complex network, the digital images acquired by being sampled and quantified need to be transformed into descriptive forms of complex network. Once an image is represented as a complex network, it becomes possible to devise more complicated algorithms for its characterization and analysis which rely not only on local properties (i.e. pixels and respective neighborhoods), but also on medium to long range relationships. For instance, the node degree, as well as many other measurements associated to individual nodes (e.g. clustering coefficient and hierarchical measurements) can be used as subsidy for identification of textures and even objects. Based on the analysis of the above-mentioned, the classic methodology of the content-based image retrieval and the properties of complex network, a new algorithm, CNBIR, (i.e. complex network-based image retrieval) is put forward. According to the methodology, sub-network constituted by mutual connective nodes is obtained, and then its block characteristics are picked up and the eigenvector is accounted. Last, determination is accomplished by use of measure function of similarity in terms of experience to realize the retrieval function. Due to considering the visual information of the color and dispersion of images, as well as applying the statistics characteristics of complex network, the experiment result shows that the methodology has better effect on image retrieval. Finally, the thesis summarizes and expects the brilliant future of applying the complex network to image retrieval. In the process of constructing from digital images to complex network, the constitution ways of edges greatly affect the retrieval result. In the meantime, the selecting quality of various kinds of statistics of complex network and network communities will directly reflex the capability of image retrieval. If people can apply the newly springing up techniques of complex network to image retrieval, it is quite sure to improve the speed of image retrieval greatly.更多还原