一种采用Gabor小波的纹理特征提取方法

An Approach of Using Gabor Wavelets for Texture Feature Extraction

Gabor小波是一种重要的纹理特征提取方法。利用其基函数的正交性,Gabor小波不仅可以有效地提取纹理特征,而且可以消除冗余信息。然而,采用Gabor小波方法计算得到的纹理特征向量具有较高的维数,因此,提出一种采用Gabor小波的纹理特征提取方法。该方法采用Gabor小波方法计算不同尺度和方向的能量信息,根据这些信息确定了显著峰集合。根据显著峰集合,确定了纹理特征向量,并且把显著性作为权重引入到相似性度量。实验结果表明,采用该方法的系统具有和采用直接Gabor小波变换方法的系统近似相同的检索性能,而纹理特征向量的维数仅为采用直接Gabor小波变换方法计算得到的纹理特征向量维数的6.1%。

Gabor wavelets are one of the important approaches to texture feature extraction. Through the orthogonality of its base functions, the Gabor wavelets can not only extract texture features effectively, but also reduce redundancy. However, the texture feature vector computed from the Gabor wavelets has higher dimension. An approach using modified Gabor wave- lets is presented in the paper. The approach uses the Gabor wavelets to compute energy of different scales and different directions, and the dominant peak set. Then the texture feature vector is computed from the dominant peak set. Furthermore, standardized energy is introduced into similarity measure as weights. Experiments show that the system that uses the modi- fied Gabor wavelets has about the same retrieval performance as that uses the Gabor wavelets. However, the dimension of the texture feature vector of the former is only 6.1% of that of the latter.