论文部分内容阅读
目的 Snakes模型对曲线轮廓具有良好的拟合能力,被广泛应用于遥感图像的道路提取。但SAR图像受乘性斑点噪声影响严重,因此利用Snakes模型从SAR图像提取道路时,传统的以图像灰度负梯度为外部能量的方法难以取得理想结果。针对这一问题,利用计算机视觉中的张量投票算法可以从噪声掩盖的图像中提取显著结构特征的特点,将张量投票与Snakes模型结合从SAR图像提取道路。方法首先利用模糊C均值分割法从SAR图像中分割出道路类,然后对道路类进行张量投票获得每点的曲线显著性值,最后以该曲线显著性值的负值作为Snakes模型外部能量从SAR图像提取道路。在Snakes模型能量最小化阶段,提出了一种优化的拟合策略,一边内插节点一边最小化Snakes模型能量。结果利用机载和星载不同场景的SAR图像进行实验,与同类的基于Snakes模型的半自动方法相比,本文方法对曲率较大的道路仅需较少控制点即可取得较好的拟合效果;与基于MRF模型的自动方法相比,本文方法对道路提取的完整率、正确率、检测质量都优于基于MRF模型的方法,并且提取的时间远远快于基于MRF模型的方法,对于大范围的道路网提取将更为实用。结论本文方法充分考虑到道路的几何形态特征,利用张量投票算法对该特征进行量化,并利用优化的拟合策略来最小化Snakes模型能量来提取道路。基于机载和星载SAR图像的实验表明本文方法可以较好地提取不同场景中的主要道路目标和道路网。
Objective Snakes model has good fitting ability to curve profile and is widely used in road extraction of remote sensing images. However, the SAR image is seriously affected by multiplicative speckle noise. Therefore, when the Snakes model is used to extract the road from the SAR image, it is difficult to obtain the ideal result by using the negative gradient of the image as the external energy. In response to this problem, the tensor voting algorithm in computer vision can extract the features of significant structural features from the noise-masked images, and combine the tensor voting with the Snakes model to extract the road from SAR images. Methods Firstly, the road classes were segmented from the SAR images by fuzzy C-means segmentation, and then the tensor voting of road classes was performed to obtain the saliency value of each point. Finally, the negative value of saliency value of this curve was used as the external energy of Snakes model SAR image extraction road. In the energy minimization phase of Snakes model, an optimized fitting strategy is proposed, which minimizes the Snakes model energy while interpolating nodes. Results Compared with the semi-automatic methods based on Snakes model, the SAR images of different scenes in airborne and on-board experiments are compared. The proposed method can achieve better fitting effect on less-curving roads with fewer control points Compared with the automatic method based on MRF model, the method proposed in this paper is superior to MRF-based method in the completeness, accuracy and detection quality of road extraction, and the extraction time is much faster than that based on MRF model. For the large Range of road network extraction will be more practical. Conclusion This method takes full consideration of the geometric features of the road, quantifies this feature by the tensor voting algorithm, and uses the optimized fitting strategy to minimize the energy of the Snakes model to extract the road. Experiments based on airborne and spaceborne SAR images show that the proposed method can extract the main road targets and road networks in different scenarios well.