Object and rule based approach for classification of high spatial resolution data over urban areas

Using the inherent features of high resolution data, such as the shape and the texture, this paper proposed an object and rule based fuzzy classification method. First, multi-scale segmentations were used to obtain homogeneous objects at different scales. According to fuzzy classification ideas, these segmented objects were further classified by using the corresponding spectral, shape, texture, topology and other object-related characteristics. This method not only overcomes the limitations of pixel based classifications, but also takes advantage of the inherent features of high resolution data. To fully compare and analyze the proposed classification method, an IKONOS image of urban areas was selected as test data. According to four main classification steps, this data was classified as houses, roads, vegetation, and bare land. The classification results showed that the proposed method enhances the accuracy of classification and is of great advantages compared with the traditional pixel based classification methods.