Research On Interactive Road Extraction Method From High-resolution Remote Sensing Images

The Remote Sensing technique developing fast recently provides abundant Remote Sensing Images with higher and higher resolution for acquiring the information of earth observation. The data translate to useful information only through interpretation. Remote Sensing image interpretation is classified into visual interpretation, interactive interpretation and automatic interpretation acoording to automation level. Because of excessive workload and low efficiency that visual interpretation suffered and unsatisfied interpretation result automatic interpretation produced due to extreme complexity of the problem under current theory and technique level, it is practical to develop human-computer interactive interpretation system. Road, as basic geographic information in the complicated and diverse Remote Sensing Images, is extremely important. So road identification and location from the images is very valuable. Recent researches focus mainly on road extraction from middle or low resolution images, but road extraction from high-resolution images is quite different and more difficult because of different image characteristics of roads. So it is urgent and important to develop road extraction algorithm for high resolution Remote Sensing Images. Under this circumstance, a Remote Sensing Image interactive interpretation system is designed and interactive road extraction methods are proposed in this paper. These methods are used in the interactive interpretation system as assistant tools. The content of this paper includes:(1) the research of preprocessing method for smoothing road images;An image smoothing method with coherence enhancing is presented after analyzing several typical nonlinear diffusion filtering methods. It extends nonlinear diffusion method to solve the paradox of enhancing structure information and smoothing homogenous areas. By selecting different diffusion matrices, isotropic and anisotropic diffusion are unified in the numerical scheme, which simplifies computation. The experiments show that it is efficient in smoothing road image and enhancing road structure information synchronously.(2) the research of interactive extraction methods for two typical roads (roads in residential area and main roads in urban) of high-resolution urban image;According to linear feature of road in residential area, linear road segment (network) extraction methods and relative technique are investigated, which include local road main direction detection method, interactive Gibbs-sampling-based road network extraction method and real-time interactive road segment (network) extraction method. The Gibbs-sampling-based method is accelerated by introducing main direction detection. And the real-time method is based on detecting main direction. The proposed robust road main direction detection method consists of initial main direction extraction based on the direction histogram of structure tensor and precise direction detection on vectorized edge image. Experiment results demonstrate its robustness and precise. The proposed Gibbs-sampling-based method models road network with marked point process and can effectively extract road network with initial configuration. However it is slow and still can not deal with single and isolated road segment. These problems can be solved by the real-time method. And experiments verify that the real-time method is fast during extracting road segment (network).According to image characteristics of main road in high-resolution urban images, an algorithm of parallel active contour model based on multi-scale frame is proposed to extract and represent main road segments. The method makes full use of the advantage of multi-scale frame and the property of active contour model to get initial location of road sides in the highest scale and localize precise positions in lower scales. The optimization computing course unifies location and denotation process. Experiment results indicate that algorithm's effectiveness in extracting main road segents from high-resolution images. The method can extract curved main road and dispose of partial shaded problem on roads.(3) the design of the interactive interpretation system for Remote Sensing Image;The designed interactive interpretation system proffers interpreter a good framework with good performance in interactiveness, compatibility and extendablity. The experiment of using comprihensive methods to interprete road image validates the point that the proposed methods could be powerful assistant tools for interpreter and promote the efficiency of image interpretation.