| Extraction of buildings in remote sensing images is a classic topic.Buildings as the city’s core functional facilities,its detailed grasp can help humans to make cities better.Traditional extraction of buildings is based on pixels,that is,whether each pixel in the image is a building part.However,the boundaries of buildings thus extracted may be inaccurate,and due to the different conditions of each image,the difference of the extraction results on different images will be larger.With the increase of the resolution of remote sensing images,object-oriented extraction methods are more and more important.Object-oriented methods require analysis of shape features.The traditional shape features are rectangles,asymmetry,but these are not very good extraction features for buildings.This paper proposes an effective object-oriented building extraction method that combines a new shape feature index.The specific implementation method is as follows:firstly,an appropriate segmentation method for the image is selected to perform effective segmentation to form a target object.An effective shape feature is designed and combined with the spectral information of the target object,and a support vector machine classifier is used to perform the classification extraction of the building.This article focuses on object-oriented building extraction and mainly does the following research work:(1)The classic pixel-based building extraction method was analyzed.The common gray level co-occurrence matrix and morphological building index were analyzed.The study found that the pixel-based extraction method does have the characteristics of unclear boundaries and ambiguous extraction and classification results,and summarizes these shortcomings.(2)An improved MRF image segmentation method is proposed.Segmentation algorithms such as watersheds and active contour models are poorly automated and require manual provision of initial values(seed points and contours),at the same time traditional MRF segmentation is too sensitive to noise.Experimental results show that the improved MRF segmentation algorithm retains more detail and its noise immunity is also better.(3)A new type of rule index describing the shape of the target is proposed.The index describes the degree of regularity of the target shape and is represented by numerical values.Artificially constructed buildings have the characteristics of straight outline rules.The final experimental results show that the use of this index combined with spectral information,and using the support vector machine classifier for classification and extraction,the overall accuracy of more than 90%. |
