• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Asian Conference on Remote Sensing(ACRS2005); Asian Space Conference; 20051107-11; 20051107-11; Ha Noi(VN); Ha Noi(VN) >Improving Classification Accuracy by Combining Spectral and Texture-based Feature Spaces in High Resolution Multispectral Images
【24h】

Improving Classification Accuracy by Combining Spectral and Texture-based Feature Spaces in High Resolution Multispectral Images

机译：通过组合高分辨率多光谱图像中基于光谱和纹理的特征空间来提高分类精度

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Although conventional classification methods based on the use of spectral information may provide acceptable results on low and medium resolution images, usually they have shown poor results when applied for high resolution images. Many researchers have shown that contextual information is a rich source of information for improvement of the classification accuracy. In this paper, texture quantization as an example of producing valuable features for the purpose of object discrimination from IKONOS pan-sharpened data of suburb areas has been investigated. Several statistical features such as the mean, variance and median as well as the features originated from the gray level run-lengths matrix have been generated. In addition, autocorrelation and geo-statistical methods have been employed for production of the new features. Results of different tests have shown that by proper use and combination of texture-based features in the classification process, up to 20 percent improvement in the accuracy may be achieved
机译：尽管基于光谱信息使用的常规分类方法可以在低分辨率和中分辨率图像上提供可接受的结果，但是当应用于高分辨率图像时，它们通常显示出较差的结果。许多研究人员表明，上下文信息是提高分类准确性的丰富信息来源。本文以纹理量化为例，以从IKONOS郊区锐化的郊区数据中识别物体为目的，研究了产生量化特征的示例。已经生成了一些统计特征，例如均值，方差和中位数以及源自灰度级游程长度矩阵的特征。另外，自相关和地统计方法已被用于产生新特征。不同测试的结果表明，通过在分类过程中正确使用和组合基于纹理的特征，可以将准确性提高多达20％

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号