Image classification with spectral and texture features based on SVM

In this paper, a three-step classification method is proposed for remote sensing images with the spectral and texture features based on the Support Vector Machine (SVM) classifier. The image is first segmented into regions with the spectral features. Then, texture features are extracted from each region by the undecimated wavelet transform. Third, the SVM is used to classify the image with these extracted texture features. A postprocessing method is also proposed to handle the small regions and anomalistic regions.