遥感影像线状地物智能化提取算法研究

对遥感影像线状地物的提取进行了深入研究分析,分别采用Snakes模型、二值形态学和面向对象处理等3种算法智能化提取线状地物。通过实验验证,3种智能算法都是行之有效的,且Snakes模型在各种线状地物提取上表现稳定,而二值形态学和面向对象算法在部分线状地物提取上表现较为突出,整体稳定性却不如Snakes模型。     

改进角度纹理特征提取高分辨率遥感影像带状道路

针对目前高分辨率遥感影像的道路自动提取算法研究中的不足,该文提出了一种基于并行角度纹理特征的半自动道路提取算法:用户输入完成道路中心线上的起始点、道路方向、道路宽度等初始化工作,利用并行角度纹理特征获取道路前进方向,用抛物线参数方程构建道路轨迹模型来预测道路轨迹点,使用角度纹理特征值构建的紧质度系数和抛物线的曲率变化来约束道路轨迹点,验证失败则转入手工跟踪;往复执行以提取道路中心线。试验证明,本算法是一种稳健的道路半自动提取算法。     

Rectilinear building roof contour extraction based on snakes and dynamic programming

This paper presents a method for extracting building roof contours from digital images collected over urban landscapes. The proposed method utilizes an energy function based on snakes that represents building roof contours in digital images and is optimized with a dynamic programming (DP) algorithm. Because most building roof contours are characterized by rectilinear sides that intercept at right angles, appropriate geometric constraints are enforced in the previously reported snake-based energy function. The main advantage of using the DP algorithm for optimizing the proposed snake-based energy function is its better radius of convergence compared to that typically obtained in the original solution based on variational approaches. Experimental evaluation, which included visual inspections and numerical analyses, was performed using real data, and the obtained results demonstrated that the proposed method has significant potential for successfully extracting building roof contours from digital images.     

Global and local methods for tracking the intersection curve between two surfaces

At present, the modelling of terrain edges from discrete data clouds {x,y,z} is one of the ‘hot topics’ in the processing of laser scanning data. This paper proposes two different methods for the three-dimensional modelling of terrain edges. Common to both methods is the idea to describe the terrain edge as the intersection line of two surface patches z_i=z(x,y), i=1,2. The first method is based on numerical integration of a differential equation describing the intersection line. The second method uses the snakes algorithm for the identification of the terrain edge. Both methods are tested for synthetic and real-world data examples, which shows that they are suitable for the practical extraction of edges from laser scanning data.     

一种复杂背景下的人脸特征提取算法

通过人脸特征提取是计算机人脸识别技术中的核心技术。主动轮廓模型方法(Activecontourmodel,也称Snakes方法)被广泛应用于特征提取、图像分割等技术中。常规Snakes方法应用于人脸特征提取鲁棒性不强,并且精度不高。对常规Snakes方法进行了改进,其改进主要包括两个方面:1)针对人脸肤色的特点改造了Snakes能量函数的图像能量项,用以提高算法的精度;2)基于人脸五官形状特点,增加了一个新的能量项———形状限定能量项,以期获得更好的鲁棒性。最后,对改进前后的Snakes方法进行了实验仿真与比较、对比研究。实验结果表明,改进后的Snakes方法具有良好的收敛性,并能准确收敛到人脸特征处。     

Modelling of irregularly sampled surfaces by two-dimensional snakes

A new method for the modelling of irregularly sampled surfaces is presented. The surfaces under consideration allow discontinuities of the first and higher derivatives. The method discussed is a generalization of the so called snakes approximation method. The variational problem, which is part of the snakes modelling, turns out to be the weighted sum of a membrane and a thin-plate-spline kernel. The variational problem is solved by the direct method using scaled and shifted Gaussians as base functions. The efficiency of the proposed method is demonstrated for a set of laser-scanning data. AcknowledgementsThe authors would like to thank Prof. Siegfried Meier from Dresden University of Technology, Germany, for his encouragement and suggestions to this work. They also wish to acknowledge the support of the KBN, State Committee for Scientific Research, Poland under grant no. 5 T12E 057 22 and the reviewers for their helpful comments.     

基于遥感影像和矢量数据的水系变化检测

介绍了基于遥感影像和矢量数据的水系变化检测方法。根据矢量的拓扑结构,将水系分为主干河流、湖泊和支流水系网。对主干河流采用"模式分类、区域生长、形态学处理、边缘提取、缓冲区检测"的流程,对湖泊采用"区域生长、形态学处理、多边形面积相减"的处理,对支流水系网采用"金字塔分解和多尺度模板匹配、Ziplock Snakes模型优化、缓冲区检测"的策略。实验结果验证了上述方案的可行性。