A Fractional Extraction Algorithm Of Building Features From High Resolution Remote Sensing Images

The electromagnetic wave characteristic of ground objects from Remote Sensing data comprehensively reflects the spectral, shape, texture information. With the development of high-resolution remote sensing technology, the subtle changes of the surface features can be obtained from the image of remote sensing. As an important element of city, township and rural areas, extraction and identification of buildings have important application value and research significance. Considering the building with complex characteristics, processing method based on gray feature traditional has many disadvantages of slow processing speed, no-automation, limited extraction accuracy. It is essential to study the building extraction method based on high resolution remote sensing image, especially for update spatial data, urban planning and image understanding. In current, automatic extraction of buildings was the hotspot in the domain of Remote Sensing theory and application.Based on the introduction of the development status of high resolution remote sensing data and the method on building feature extraction, the spectral, shape and texture feature of buildings in the research area were analyzed. A sequential extraction method for the buildings is put forward through comparing the several typical segmentation methods. This dissertation includes the following contents.According to the noncontinuity characteristics of the brightness value in building edge, the segmentation method is studied based on edge detection, using the Canny operator in the experiment their characteristics were analyzed. Considering the heterogeneity of buildings spectra, segmentation method based on region is adopted, and regional growth algorithm and Otsu segmentation method were analyzed. Building information were extracted by combination the former methods. Discusses the segmentation principle based on clustering, ISODATA and EM algorithm is analyzed, which prepare for subsequent classification by experiment. A sequential extraction algorithm is presented based on the analysis of the characteristics of the building.(1) Non-building information is removed according to the building spectral heterogeneity, segmentation methods based on region growing and threshold combining to the initial segmentation of image.(2)In order to remove the vegetation information with spectral average, the study adopts the vegetation index using the spectral radiation difference of buildings and vegetation.(3) According to the morphological structure of road shows in the image, some road information, which can interfere the characteristics of building, is removed by morphological method of image processing.(4) According to rich texture information provided by the image, this study established feature vector using the Log-Gabor filter to extraction texture information. The experiment is carried out by using the model Bias classification rules to separate buildings and non-building, which accuracy rate is91.07%. The results show that our proposed method is effective to improve the degree of automation of building extraction, and it has the certain reference value to the high resolution remote sensing image processing, analysis and application.