基于NSST和MCM纹理特征提取算法  

Texture Feature Extraction Algorithm Based on Nonsampled Sheartlet Transform and Multiresolution Co-occurrence Matrix

作　　者：魏远远[1] 殷明[1] 

机构地区：[1]合肥工业大学数学学院,合肥230009

出　　处：《大学数学》2016年第1期15-25,共11页College Mathematics

基　　金：国家自然科学基金(11172086);安徽省自然科学基金(1308085MA09)

摘　　要：纹理特征提取作为图像处理的重要环节,对图像的后续处理有着至关重要的影响.文中在多分辨共生矩阵算法的基础上,针对标准Brodatz纹理图像检索,通过非下采样剪切波变换的多分辨共生矩阵和混合高斯模型相结合,提出了一种纹理特征提取算法.文中首先对Brodatz纹理图像进行非下采样剪切波变换得到子带系数,通过对细节子带直方图分析,引入了拟合效果较好的混合高斯模型.然后利用优化的非均匀量化策略,提取多分辨共生矩阵纹理特征F2和F10.最后将提取的纹理特征与统计特征级联融合并结合具有权重系数的相似性度量公式,用于最终纹理图像检索.仿真实验表明:与传统多分辨共生矩阵的方法相比,文中所提算法的平均检索率分别提高了2.01%和8.87%.Texture feature extraction as the important element of image processing has critical influence for the subsequent processing of the image. On the basis of multi-resolution co-occurrence matrix（MCM）, by combining the MCM of nonsampled shearlet transform（NSST） with gaussian mixture model（GMM） come up with a novel texture feature extraction algorithm for texture retrival of the standard Brodatz. The algorithm in this thesis obtained subband coefficients from NSST at first, and introduced GMM by analysising the detail subband histogram. Then extracted the texture feature F2 and F10 by making up with the optimal nonuniform quantizing. At last fused the texture feature of MCM with the statistical feature, and integrated similarity measure formula with weight coefficient for texture image retrival. The simulation results shows： compared with the traditional method of MCM, the average retrival rate of the algorithm increased by 2.01 % and 8.87 % respectively.

关 键 词：多分辨共生矩阵 非下采样剪切波 混合高斯模型 非均匀量化 纹理特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]