期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2008
卷号：XXXVII Part B7
页码：707-712
出版社：Copernicus Publications
摘要：Texture quantization is a useful method for extraction spatial relevance between pixels which is used in humane brain for image interpreting. Beside the spectral bands textural features of high spatial resolution data can be used to improve classification accuracy. Depends on the land cover characteristics different textural features possibly are effective from large number of available textural features. So it is important to find proper features among available features for special case studies. In this paper efficient features are determined by ranking based on their ability for improving class separability. The quadratic discriminate classifier (QDC) and support vector machine (SVM) are used for data classification. Comparative tests on compact of texture features and training sample size to improve accuracy of QDC and SVM classifiers demonstrated that i) QDC is an efficient classifier with large number of training samples while due to using more texture features led to futile result in high dimensional feature space; ii) SVM generates accurate results in high dimensional feature spaces and can train with few training samples. Experimental results show 13% and 10% improvement in obtained average and overall accuracies respectively
