|
|
|
Improving Land-use Mapping in Complex Landscapes by Combining Spectral and Texture-based Feature Spaces
Hamed Ashoori
Student
K.N. Toosi University of Technology,
Iran
Email: h_ashouri@sina.kntu.ac.ir
Classification is the most common method for land use mapping using remotely sensed images. Customary classification methods use only spectral information which doesn't result in very accurate results, but combining other features extracted from spectral bands could improve the results accuracy. Especially in high spatial resolution images poor results are obtained through spectral classification. To extra the features mentioned above try to quantify existing relations between adjacent pixels. In this paper to extract these features some available methods such as first order statistical, gray level co-occurrence based and fourier based methods, were evaluated. These methods have different effects on outputs accuracy so considering different aims of classification, such as obtaining the best results in some special classes or generating a homogeneous map, different combinations of features should be applied. As a result the mentioned methods were classified and then proper combinations to achieve different purposes are proposed.
|
|
|