轮廓波域Entropy-Fourier旋转不变算子在纹理分类中的应用  

Rotation Invariant Texture Classification Based on Entropy-fourier Descriptor in Contourlet Domain

作　　者：冯毅[1] 许灿辉[2] 杨家红[1] 钟坚成[1] 

机构地区：[1]湖南师范大学工学院,湖南长沙410081 [2]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《计算技术与自动化》2008年第3期16-19,共4页Computing Technology and Automation

基　　金：湖南省自然科学资金资助项目(08JJ3131)

摘　　要：针对图像的旋转问题,提出一种基于contourlet域的旋转不变纹理描述算子。对contourlet变换后的系数进行处理,提取子带香农熵作为特征向量,然后对各个尺度的特征向量进行离散傅立叶变换,得到旋转不变的纹理特征。利用傅立叶谱的对称性,将特征向量进一步降维。最后采用支持向量机和欧式距离对纹理图像进行分类。在包含1456幅图像的自建图像库中进行分类实验,结果表明该方法在低计算复杂度的优势下,能获得较好的分类性能。This paper presents a rotation invariant texture classification method. The new method extracts a set of Shannon entropy in contourlet domain. Discrete Fourier Transform analysis is then applied to entropy vectors of each scale to form rotation invariant feature vectors, which are reduced further due to the symmetry of DFT magnitude spectrum. Two classifiers, including the Euclidean distance and Support Vector Machine, are studied to measure the distance between the known and unknown features. Experimental result on 1456 texture images has demonstrated that contourlet is an efficient tool to represent directional texture, and the proposed rotation invariant method can achieve accurate classification with low computational complexity.

关 键 词：CONTOURLET 香农熵 离散傅立叶变换 支持向量机 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]