摘要
采用完全树式结构小波包分解方法提取了声纳图像的纹理特征。通过计算小波包系数的统计特征和共生矩阵,构造了一个基于小波包变换的统计—共生矩阵特征集。由于统计—共生矩阵作为图像纹理特征表示时,在保留图像统计特征的同时,引入了空间信息,所以识别性能明显优于单纯的统计量识别方法。最后设计了一种模糊纹理分类器,由于模糊判决分类器通过引入隶属度函数对特征进行模糊化,反映了各类纹理样本间由于随机噪声等畸变因素造成的抽取特征值存在的不确定性,从而提高了纹理分类器的鲁棒性。
The texture features of sonar images are extracted by tree structured wavelet packet decomposition. A statistical and co-occurrence matrix feature set is worked out by calculating statistical feature and co-occurrence matrix of wavelet packet coefficients. As the texture features, statistical and co-occurrence matrix not only contains the statistical features but also includes spatial information of images. So the classification of it is much better than statistics methods. A fuzzy texture classifier is designed. By introducing membership function of the feature, the classifier reflects the uncertainty of the feature values made by random noise. This increases the robustness of the texture classifier.
出处
《系统仿真学报》
CAS
CSCD
2004年第8期1673-1675,共3页
Journal of System Simulation
基金
"十五"预研项目(1530501200017)
关键词
小波包
树式结构
统计—共生矩阵
模糊分类器
wavelet packet
tree structure
statistical and co-occurrence matrix
fuzzy classifier
