High-resolution Remote Sensing Images Of Urban Road Extraction Based On Mathematical Morphology

In recent years, as the rapid development of remote sensing technology, satellite remote sensing image resolution is also improved significantly, and extracting the road information from the high resolution image has many advantages such as short cycle time, fast speed and rich detail. The city road network of remote sensing images is not only a very important component of geographic information, and also the basis applications data of geographic information system. Extracting the road information has far-reaching significance in GIS data acquisition, remote sensing image understanding, mapping and spatial database updates, etc. However, the existing method of extract road from remote sensing images doesn't have ideal effect, or mistaking the road goal as the background so as to omit objectives, or mistaking the background as the road target. As a result, how to automatic processing, interpreting massive amounts of image data, and use these data to achieve the dynamic update GIS data has become an important issues in the process of information technology society.The Mathematical morphology in the image processing has some characteristics such as simplifying image data, maintaining the basic shape of the image characteristics, removing irrelevant structures and simple, flexible and fast operation and so on. In order to improve the accuracy of road extracting,the paper emphatically improves the effect of image segmentation, introducing the image's texture and Spectral feature and proposing a method about clustering segmentation based on an synthesized feature. The synthesized feature is made up of three kinds of texture features and two kinds of spectral features. Though choosing the moving four parameters including window's size, distance, gray level and relative direction of pixel pairs, making use of Gray Level Co-occurrence Matrix to extract image texture feature, comparing and analyzing five texture features that is suitable for remote sensing image, determine contrast, angular second moment and entropy as the texture feature. And choose gray-scale mean and standard deviation of the image pixel as the spectral features.After the image segmentation based on synthesized feature, the paper makes use of the mathematical morphology opening operation,morphological reconstruction, morphology closing operation to separate the image of roads and buildings which adhere the road, get rid of the non-road noise, Connect abrupt road ,and then fill the image road holes, morphological thinning, remove burr etc. Finally, we get result of the road extraction.Experimental results show that clustering segmentation method based on an integrated feature of the road obtains better segmentation effect than image segmentation based on single feature. Through visual overlaying and quantitative analyzing, we judge the final result of road extraction has very high accuracy. As a result, the road extraction method proposed by this paper is feasible and achieved favorable results.