摘要
基于偏最小二乘回归技术对纹理特征进行线性组合,得到新的纹理特征来进行分类。实验表明,组合后的纹理特征不但提高了纹理分类的性能,而且具有一定的数据自适应能力。
The paper presents partial least squares （PLS） method. Firstly, texture features （spectrum （TS） and gray-level co-occurrence matrix （GLCM）） are calculated from local image regions. Secondly, the authors apply PLS regression to preparatory texture features to extract linear combined new texture features. Thirdly, both the linear combined texture features and the preparatory texture features, together with the ordinary texture features, are imported into linear discrimination analysis （LDA） and quadratic discrimination analysis （QDA）. Finally, classification results are compared and conclusions are drawn.
The experiments show that not only PLS can reduce the dimension of texture features but also the combined texture features efficiently have better discrimination abilities than the ordinary texture features.
出处
《武汉大学学报（信息科学版）》
EI
CSCD
北大核心
2006年第5期399-402,共4页
Geomatics and Information Science of Wuhan University
关键词
特征组合
特征选择
共生矩阵
纹理谱
偏最小二乘回归
feature combination
feature selection
GLCM
texture spectrum
partial least square regression
