摘要
提出了一种新的基于非下采样Contourlet变换的纹理特征提取方法。首先对纹理图像进行非下采样Contourlet变换,然后提取不同尺度、不同方向上变换系数矩阵的均值和方差作为特征向量,大大降低了特征维数,并利用BP神经网络进行训练和仿真,实现了纹理图像的自动分类。实验结果表明,与小波包变换和改进的LBP纹理算子等方法相比,该方法能取得更好的分类效果。
A novel approach for feature extraction of texture images based on NonSubsampled Contourlet Transform （NSCT） was proposed. The coefficients in different scales and different directions were obtained by textural image decomposition using NSCT. Then the means and variances of theses coefficients were extracted to be the feature vectors, which could greatly reduce the number of feature dimension. Back Propagation （BP） neural network was adopted to implement automatic classification of texture images through training and simulation. Compared with wavelet package transform and the improved Local Binary Pattern （LBP） texture descriptor, this approach can achieve better result.
出处
《计算机应用》
CSCD
北大核心
2009年第12期3434-3436,共3页
journal of Computer Applications
基金
国家自然科学基金资助项目(60872161)
泰山学院科研基金计划项目(Y06-2-16)
关键词
非下采样CONTOURLET变换
纹理分类
小波包变换
局部二值模式
NonSubsampled Contourlet Transform （NSCT）
texture classification
wavelet package transform
Local Binary Pattern （LBP）
