Context Aware Modification on the Object Based Image Analysis

Spectral similarities, spatial adjacencies and complex relationships between various object types lead to the recognition difficulties and ambiguities in complex urban areas. Using a knowledge base containing contextual information together with higher levels of modifications in the object based image analysis may improve recognition results in such a situation. In this paper, contextual information composed of regions' internal and external context are utilized for object classification and modification based on the pre-defined segmented regions in very high resolution (VHR) remotely sensed imagery. In the proposed methodology, defining region's internal context for each of the segmented regions, object based image analysis (OBIA) is performed. Then, for the modification of the wrongly classified regions due to the internal context similarities, region's external context is defined based on the mutual dependencies between neighboring regions in the image. Evaluation of the capabilities of the proposed context aware object recognition methodology is performed on the WorldView-2 satellite imagery over Rio de Janeiro in Brazil together with a digital surface model. The obtained results represent that using various contextual information together with higher levels of modifications on the object regions, some of the object recognition difficulties due to the spectral similarities, spatial adjacencies and mistakes in generating the classification rules are decreased.