基于边缘点特征的高分辨率遥感图像配准方法研究

发布时间：2018-07-16 21:39

【摘要】：遥感图像配准是后端多种遥感应用的基础和前提,配准精度直接影响后续应用的效果,配准的自动化处理和配准效率的提高对海量遥感数据分析具有重要意义,也是学者们关注的焦点。SIFT(Scale Invariant Feature Transform)和SURF(Speed Up Robust Features)是常用的典型遥感图像配准方法。SIFT具有尺度不变性,在噪声干扰、仿射变换等方面有良好的鲁棒性,但计算量大,耗时长,且采用欧氏距离匹配特征点会产生较多误匹配点对。SURF算法优化了配准时间,在尺度和仿射变换下保持不变性,但配准精度有待提高,匹配率较低。随着遥感图像分辨率的日益提高,遥感图像的尺寸和数据量不断增大,以及随着遥感应用的发展,都对图像配准的性能提出越来越高的要求,特征配准方法克服灰度配准方法的局限,是目前遥感图像配准领域研究的重点。本论文面向细节纹理信息丰富的高分辨率遥感图像,提出一种基于边缘点特征的高分辨率遥感图像自动配准方法。该方法主要包括五个方面:(1)对参考图像与待配准图像都进行一级Haar小波变换,基于小波变换后的低频近似图像进行匹配,然后根据匹配结果完成原图的配准,从而有效减少计算量,提高配准速度;(2)根据不同种类遥感图像的特性使用不同的特征提取算法(光学图像使用Canny算子,SAR图像使用Ratio Of Averages算子)来提取边缘点特征,利用边缘点特征既可准确定位,又可获得稳定的特征。(3)在特征匹配环节,同时考虑特征点的主、辅方向特征,使得一个特征点拥有多个方向的特征描述,从而增强图像配准的鲁棒性,然后根据特征点间最小角与次小角的角度比值小于某个阈值来确定初始匹配点对;(4)在匹配点对筛选环节,对随机抽样一致性算法进行改进,通过添加约束条件,优先选择质量好的匹配点对进行模型参数拟合,以此提高图像配准的精确度;(5)仿射变换环节,采取分区域选取匹配点对的策略,从而使匹配点对在图像中均匀分布,以避免配准中的局部最优问题,进一步提高图像配准精度。为验证上述方法的有效性,本论文针对不同的情况进行了实验验证:相同传感器光学图像配准、相同传感器SAR图像配准、不同波段间的图像配准、不同分辨率的图像配准以及不同卫星传感器的图像配准。并将本文方法与比较典型的SIFT算法、SURF算法进行比较分析,采用匹配率、匹配效率、均方根误差和时间消耗四个定量评价指标来客观评价算法的配准性能。实验结果表明,本论文提出的基于边缘点特征的高分辨率遥感图像自动配准方法具有较高的配准精度、较好的鲁棒性,且在处理效率方面具有较好的有效性。
[Abstract]:Remote sensing image registration is the foundation and premise of many remote sensing applications. Registration accuracy directly affects the effect of subsequent applications. The automatic processing of registration and the improvement of registration efficiency are of great significance for the analysis of massive remote sensing data. SIFT (scale variant feature transform) and SURF (Speed up robust Features) are typical remote sensing image registration methods, which have scale invariance and good robustness in noise interference and affine transformation. Using Euclidean distance matching feature points can produce more mismatch points to optimize the registration time of SURF algorithm and keep the invariance under the scale and affine transformation but the registration accuracy needs to be improved and the matching rate is low. With the increasing resolution of remote sensing images, the size and data volume of remote sensing images are increasing, and with the development of remote sensing applications, the performance of image registration is becoming more and more demanding. Feature registration method overcomes the limitation of gray level registration method and is the focus of research in remote sensing image registration field. In this paper, a high resolution remote sensing image registration method based on edge features is proposed for high-resolution remote sensing images with rich detail texture information. The method mainly includes five aspects: (1) Haar wavelet transform is applied to both the reference image and the image to be registered, and the low-frequency approximate image is matched based on the wavelet transform, and then the original image is registered according to the matching result. In order to reduce the computation cost and improve the registration speed effectively, (2) according to the characteristics of different kinds of remote sensing images, different feature extraction algorithms are used (using ratio of Averages operator in optical images) to extract edge features. The feature of edge points can be accurately located and stable features can be obtained. (3) in feature matching, the main and auxiliary directions of feature points are considered at the same time, so that a feature point has a feature description of multiple directions. In order to enhance the robustness of image registration, and then determine the initial matching point pairs according to the ratio of the minimum angle to the sub-small angle between feature points is less than a certain threshold. (4) in the matching point selection link, the algorithm of random sampling consistency is improved. By adding constraint conditions, the matching point pairs with good quality are selected first for model parameter fitting, so as to improve the accuracy of image registration. (5) the affine transformation link is adopted to select matching point pairs in different regions. In order to avoid the problem of local optimization and improve the accuracy of image registration, the matching point pairs are uniformly distributed in the image. In order to verify the effectiveness of the above methods, experiments are carried out in this paper for different cases: optical image registration of the same sensor, SAR image registration of the same sensor, and image registration between different bands. Different resolution image registration and different satellite sensor image registration. The method is compared with the sift algorithm and the SURF algorithm. The matching rate, matching efficiency, root mean square error and time consumption are used to evaluate the registration performance objectively. The experimental results show that the proposed automatic registration method based on edge feature has higher registration accuracy, better robustness and better efficiency in processing.
【学位授予单位】：中国科学院大学(中国科学院遥感与数字地球研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【相似文献】