基于图谱理论的纹理图像分析

Texture image analysis based on graph spectral theory

图谱理论作为目前模式识别研究的热门方法之一,广泛应用于聚类与分割,但对其进行图像分析,尤其是复杂的纹理图像分析未见报道。将图谱理论引入到图像分析领域,并结合窗口纹理分析方法构造新的纹理分析算法。首先对图像灰度级进行窗口划分得到不同灰度级下的子图像,然后以各子图像作为图的顶点,子图像间的相似度作为图的边,将原图像解析为一幅带权无向图。利用图谱理论的相关思想对该无向图进行分析,可以从中获得纹理粗糙度等特征,从而完成对原始图像的分析。通过对Brodatz图像库的检索实验证明,该方法优于传统纹理分析算法。

As an active topic in pattern recognition, the graph spectral is applied in clustering and segmentation. But the analysis to image, especially the texture image, can not been retrieved till now. The graph spectral theory is introduced into the field of texture image analysis. The new method combines the graph spectral theory with windows texture analysis. The windows the image gray level is segmented at first to obtain sub-images. The sub-images are taken ss the vertexes and the similarity between them are regarded as edges of a graph. In this way, a weighted undirected graph is obtained from the original image and features are calculated from that by spectral theory. The retrieval experiments based on Brodatz dataset show that the method is superior to traditional texture analysis.