NSCT域的多尺度景象匹配方法研究

发布时间：2019-06-16 13:26

【摘要】：景象匹配是一种依靠传感器、图像匹配等先进技术,对飞行器进行精确定位的辅助导航技术。景象匹配指的是将一个图像区域从同一场景的的其他设备得到的区域中定位所在位置或找到它们之间变换参数的一种重要的图像分析和处理技术。其图像匹配技术在导航制导的末段使用,增加匹配精度,加强制导系统的自主性。目前,景象匹配技术在遥感图像处理、机器视觉、导弹制导、医学等领域有很多的应用。随着国家对航空和航天技术投入,景象匹配的性能要求匹配精度逐步提高,实时性逐渐加强;同时导航系统还应满足飞行器高机动性、全天时工作等更高的性能要求。因此高匹配精度、实时抗干扰的匹配技术是近年来研究的重点。本文主要研究了多尺度分析的图像匹配算法。利用非下采样轮廓波变换(Non-subsampled Contourlet Transform,NSCT)作为多分辨率分析的工具,针对红外和可见光图像的成像特点,对异源图像的景象匹配问题进行了研究,并提出相应算法。1.在研究了传统的降噪增强算法的基础上,针对红外图像成像质量差等问题,本章提出了一种NSCT域和空域联合的红外图像降噪和增强方法。该方法首先将算法空间变换到NSCT域,对包含噪声的图像系数按照3?准则进行处理,实现NSCT域红外图像降噪。然后在空域建立的归一化非完全贝塔灰度变换模型,实现低分辨率下低频图像增强。该方法能够完整地保留图像轮廓特征,达到增强图像质量的效果。2.在研究了多尺度图像分析理论的基础上,针对以可见光图像作匹配基准图、红外图像作实时图的问题,基于多尺度分析的景象匹配算法,提出多尺度NSCT域Krawtchouk不变矩特征的景象匹配算法。从特征空间、搜索空间和搜索策略三个方面对匹配算法进行改进,将不变矩引入特征空间利用其良好的不变性提取不变矩特征参与匹配,将特征的搜索空间带入NSCT域进一步压缩搜索空间,并利用改进的遗传算法提高搜索速度。实验表明本算法不仅具有更高的精度和速度,而且鲁棒性好。3.利用NSCT系数的稀疏性、多尺度性和各向异性,SIFT特征在图像描述上良好的的稳定性,针对景象匹配改进SIFT特征的提取,提出一种结合NSCT与改进的SIFT的景象匹配算法。该算法在NSCT域下提取改进的SIFT描述符,从搜索空间和特征空间两方面增强算法的鲁棒性,实验表明该算法能抵抗一定的噪声干扰和几何畸变。本文研究内容对于解决景象匹配中高精度,高可靠性的需求,提供了一种崭新的思路,具有一定的实用价值,可应用于可见光图像作为匹配基准图,实时图为红外图像的景象匹配问题。
[Abstract]:Scene matching is an auxiliary navigation technology which relies on advanced technologies such as sensors and image matching to accurately locate aircraft. Scene matching refers to an important image analysis and processing technique in which an image region is located from an area obtained by other devices in the same scene or where the parameters are transformed between them. The image matching technology is used in the last stage of navigation guidance to increase the matching accuracy and strengthen the autonomy of the guidance system. At present, scene matching technology has many applications in remote sensing image processing, machine vision, missile guidance, medicine and other fields. With the national investment in aviation and aerospace technology, the performance of scene matching requires that the matching accuracy is gradually improved and the real-time performance is gradually strengthened. At the same time, the navigation system should also meet the higher performance requirements of aircraft, such as high maneuverability, all-day work and so on. Therefore, the matching technology of high matching accuracy and real-time anti-interference is the focus of research in recent years. In this paper, the image matching algorithm of multi-scale analysis is studied. Using non-downsampled profile transform (Non-subsampled Contourlet Transform,NSCT) as a tool of multi-resolution analysis, the scene matching problem of heterogeneous images is studied according to the imaging characteristics of infrared and visible images, and the corresponding algorithm is proposed. On the basis of studying the traditional noise reduction and enhancement algorithm, aiming at the poor imaging quality of infrared image, a joint infrared image denoising and enhancement method in NSCT domain and spatial domain is proposed in this chapter. Firstly, the algorithm space is transformed into NSCT domain, and the image coefficients containing noise are 3? The criterion is processed to realize infrared image noise reduction in NSCT domain. Then the normalized incomplete beta gray transformation model is established in the spatial domain to realize the low frequency image enhancement at low resolution. This method can completely preserve the image outline features and achieve the effect of enhancing image quality. 2. On the basis of studying the theory of multi-scale image analysis, aiming at the problem of using visible image as matching reference map and infrared image as real-time image, based on the scene matching algorithm of multi-scale analysis, a scene matching algorithm based on multi-scale Krawtchouk invariant moment feature in multi-scale NSCT domain is proposed. The matching algorithm is improved from three aspects: feature space, search space and search strategy. The invariant moment is introduced into the feature space to extract the invariant moment feature to participate in the matching, the feature search space is brought into the NSCT domain to further compress the search space, and the improved genetic algorithm is used to improve the search speed. The experimental results show that the algorithm not only has higher accuracy and speed, but also has good robustness. Based on the sparsity, multi-scale and anisotropy of NSCT coefficients and the good stability of SIFT features in image description, a scene matching algorithm combining NSCT with improved SIFT is proposed to improve the extraction of SIFT features for scene matching. The improved SIFT descriptors are extracted in the NSCT domain to enhance the robustness of the algorithm from two aspects of search space and feature space. Experiments show that the algorithm can resist certain noise interference and geometric distortion. The research content of this paper provides a new idea to solve the demand of high precision and high reliability in scene matching, and has certain practical value. It can be applied to the scene matching problem of visible image as matching reference map and real-time image as infrared image.
【学位授予单位】：河南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】