基于Contourlet变换和支持向量机的纹理识别方法

摘要

针对变换域中图像纹理识别时如何选择最佳特征向量的问题,利用Contourlet变换的多方向、多尺度选择性和各向异性,将图像从空间域变换到频率域,全面地提取了Contourlet变换分解后低频子带、中频子带和高频子带的特征,输入支持向量机(SVM)分类器进行分类识别.利用Brodatz纹理库进行仿真实验,实验结果表明低频均值方差和高频能量作为组合特征时识别准确率可达98.75％,且特征向量维数少,是在Contourlet变换下表示图像纹理的最优特征.%How to identify the most appropriate feature vector is the key of image texture recognition. Considering the characteristics of Contourlet transform, the image was transformed from the spatial domain to the frequency domain. The feature vectors of low-frequency subband, medium-frequency subband and high-frequency subband were extracted comprehensively and entered to Support Vector Machine (SVM) for classification. Brodatz database was used for simulation. The experimental results demonstrate that mean and variance of low frequency and the energy of high frequency are the optimal representation of the image texture. They are combined to make the recognition accuracy rate up to 98. 75% and the vector dimension is lower.