双树复小波域共生矩阵的纹理特征提取方法

TEXTURE FEATURE EXTRACTION METHOD WITH DUAL-TREE COMPLEX WAVELET DOMAIN CO-OCCURRENCE MATRIX

提取纹理特征一直是纹理分析的首要问题。提出一种双树复小波域共生矩阵的纹理特征提取方法。利用双树复小波对纹理图像进行多层分解所得的低频子带图像,计算不同方向的共生矩阵,提取图像纹理特征值。实验结果表明,该方法能有效提取出多尺度、多方向的纹理特征,并兼顾纹理局部随机性和整体规律性,所提取的纹理特征具有良好的聚类分离度和类内样本差异性。

Extracting texture feature is always the primary problem in texture analysis. A new texture feature extraction method with dualtree complex wavelet domain co-occurrence matrix is proposed. The method uses the low-frequency sub-band images produced by muhilayered decomposition upon textured images by dual-tree complex wavelets to calculate co-occurrence matrixes in different directions in order to extract image texture feature values. Experimental results show that the method can effectively extract multi-scaled and multi-directional texture features, while taking into account the textures＇ local randomness as well as their overal regularity. The extracted texture features own both favorable cluster separability and sample differenciation inside a cluster.