一种快速均匀的采用尺度不变特征变换描述符进行基于内容的卫星图像配准方法（英文）

发布时间：2018-05-15 20:26

本文选题：基于内容的卫星图像配准 + 特征点分布　；参考：《Frontiers of Information Technology & Electronic Engineering》2017年08期

【摘要】：基于内容的卫星图像配准是在遥感和图像处理领域的一大难题。受照度、旋转、来源差异的影响,该问题在多源遥感图像匹配中更为突出。尺度不变特征变换(scale-invariant feature transform,SIFT)算法是一种成功应用于卫星图像配准的算法。本地SIFT描述符被许多研究者应用于改进图像检索流程。尽管SIFT算法具有良好的稳定性,它在提取多源遥感中本地特征点的质量和数量上仍然具有一定的劣势。另外,SIFT算法提取的本地特征具有较高维度,导致计算过程耗时过长以及对保存相关信息的储存空间要求过高,而这两点也是在基于内容图像检索(content-based image retrieval,CBIR)的相关应用中的重要因素。本文介绍了一种在多源遥感中将本地SIFT特征转变为全局特征的新方法。通过在预处理阶段对图像进行对比度均衡化来提升SIFT本地特征点质量和数量。将参考数据库中每副图像的本地特征单独分为一类后,采用线性判别分析(linear discriminant analysis,LDA)方法将本地SIFT特征转变为全局特征,同时不为降低特征空间的维度。该方法可以显著减少计算时间和所需存储空间。将核函数应用于检定数据并映射,所测试特征点的检索率高达91.67%。
[Abstract]:Content-based satellite image registration is a difficult problem in the field of remote sensing and image processing. Affected by illumination, rotation and source difference, this problem is more prominent in multi-source remote sensing image matching. Scale-invariant feature transform sift (scale-invariant feature transform) algorithm is a successful algorithm for satellite image registration. Local SIFT descriptors are used by many researchers to improve the image retrieval process. Although the SIFT algorithm has good stability, it still has some disadvantages in the quality and quantity of local feature points extracted from multi-source remote sensing. In addition, the local features extracted by sift algorithm have a higher dimension, which results in a long time consuming and too much storage space for storing relevant information. These two points are also important factors in content-based image retrieval (CBIR) applications. In this paper, a new method for converting local SIFT features into global features in multi-source remote sensing is introduced. In order to improve the quality and quantity of SIFT local feature points, the contrast equalization of the image is carried out in the preprocessing stage. After the local features of each image in the reference database are divided into one group, linear discriminant analysis (linear discriminant analysis) method is used to transform the local SIFT features into global features without reducing the dimension of the feature space. This method can significantly reduce the computational time and the required storage space. The kernel function is applied to the verification data and mapped, and the retrieval rate of the feature points tested is as high as 91.67.
【作者单位】： Electrical
【分类号】：TP391.41

【相似文献】