Road Information Extraction Technology Research In High-resolution Remote Sensing Images

Roads, as part of the terrain, play an important role in life. Road information extraction of theremote sensing image processing is one of the most difficult technologies; the traditionalalgorithm of road extraction was mostly aimed at low-middle resolution aviation image. With thedevelopment of remote sensing technology, the resolution of the image has been greatlyimproved; especially the low-altitude UAV technology development, the resolution of remotesensing image is improved by centimeter level which has significant edge and rich texturefeatures and would cause interference to road information extraction from sub-meter which is infavor of road information extraction. Because of the data changes, it is necessary for roadextraction technology of centimeter level to further research. The main innovations of this paperare summarized as follows:(1)The advantages and disadvantages of road extraction algorithm of image are analyzed athome and abroad, the features of high resolution remote sensing image and difficulties whichwill be confronted. In order to extract the road zone, a new hybrid-filter algorithm is produced tosmooth the image by analyzing the disadvantages of median filter and mean filter. The filteredimage road area has a more consistent gray feature.(2)A histogram peak finding technique is presented to form pattern classes for clustering,then using the modified mountain cluster algorithm to automatically get cluster number andcluster centers that is used for image segmentation. The algorithm makes full use of simplicity ofthe histogram technique and advantage of mountain cluster that automatic predict clusteringcenter, such combination effectively reduces the time of image segmentation. Then theadvantages and disadvantages of three main mountain cluster versions are analyzed andevaluated, meanwhile, as histogram technique ignore pixel spatial location, this paper try tosegment image by joining the space distance to mountain clustering algorithm.(3)Eliminating noise algorithms are setted on the basis of road characteristic in image and arediscussed and analyzed. Firstly the paper eliminates noises that were segmented in road by usingsmaller difference road gray value of high resolution remote sensing image in three colorcomponents; secondly eliminating the road shadow noise by using the characteristic indifferences of the RGB image road shadow in three components; finally eliminating the noisesthat are not connected with road structure by building morphological filter and smoothing roadboundary by using the theory of mathematical morphology. It can be seen from the road image after noise elimination that this paper’s algorithm caneffectively extract road zone in image.