SVM-based Multi-textural Image Classification and its Uncertainty Analysis

Texture analysis, a hot issue in image processing, is a key technique for ground surface object recognition. This paper presents a supervised image classification method based on multiple and multi-scale texture features and support vector machines (SVM). By taking different scales of ground surface features into account, and by feature fusion technique, this method integrates seven-dimensional texture features of different characteristics from GLCM and fractal theory to realize the land use/cover classification. The seven features combine the abilities to describe image textures of different approaches, which can reach better classification performance than any of them and significantly improves the precision of automatic image interpretation. Classification uncertainty is also evaluated and analyzed at the scale of pixel using the extended probability vector and probability entropy model. The imagery used in this research is RADARSAT-1 SAR data.