基于空间映射复Directionlet变换的图像纹理分类

Directionlet变换具有多方向各向异性基函数,能有效捕捉图像的奇异性特征.该文在此基础上构造了一种空间映射的复Directionlet变换,使其具备了更为灵活的方向选择性和近似的平移不变性.利用空间映射方法获得Directionlet变换的复函数空间,对多尺度各方向子带系数提取能量特征用于图像纹理分类.通过对Brodatz图像库及真实SAR图像的纹理分类实验表明,该文算法较之小波分析及其它多尺度几何分析方法,具有更优的纹理分类性能,也验证了Directionlet工具在图像分析中的应用潜力.

Texture Classification Based on Mapping Complex Directionlet

Directionlet transform can capture the image singularity due to possessing the multi-direction anisotropic basis functions. A texture classification algorithm based on the Mapping complex Directionlet Transform (M-DT) is proposed, which provides better directionality and approximate shift invariance. By space mapping for the texture image, then complex Directionlet transform is applied to the mapped image, and the multiscale subband coefficient energy feature is used for texture classification. The experiments using texture images from Brodatz and real SAR images indicate the proposed method outperforms wavelets and Multiscale Geometric Analysis (MGA) approaches, the potential application to image analysis by Directionlet is thus proved.