Object-oriented classification of high-resolution remote sensing image using structural feature

In this paper, an object-oriented classification method using structural feature is present. Mean-Shift algorithm is employed for multispectral band images segmentation. Straight lines are detected by Canny detection operator and Hough Transform. A new structural feature based on straight line statistics is introduced, which can be used to distinguish the artificial object and natural object in high-resolution remote sensing image. SVM is used for classification by spectral and structural features. Finally, an experiment adopting QuickBird data has been carried out to validate this method and achieved a good result.