Application of multi-level MRF using structural feature to remote sensing image classification

In order to unitize the image information at different level, this paper introduces a novel approach to integrate the pixel and region feature into MRF model. A structure feature descriptor is proposed to represent the structural characteristics of objects to disambiguate land cover types with similar spectral characteristics. The first step of the proposed algorithm is to classify the input image by using the multi-level MRF model, then the structural feature is used to classify the land cover types prone to misclassified based on the result of the first step. The proposed algorithm is evaluated by being compared with the result with single level MRF model and other existing classification method. Qualitative and quantitative experimental results show that the proposed algorithm can effectively capture the image data characteristics at different level which result in higher classification accuracy.