【作者】 蒋少华；

【导师】 王乘；

【作者基本信息】 华中科技大学 ， 空间信息系统， 2011， 博士

【摘要】 多源图像处理是目前的一个研究重点和热点,借助多源影像中包含的信息优势和互补性信息,可以生成新的更完整和更真实准确的信息,从而得到对场景中目标的更客观、更本质的认识,减少或抑制污损、残缺、伪装的目标中包含的不完全、不确定甚至错误的信息。因此目前已得到各国科研工作者的广泛关注,并已取得不少成果。但目前多源图像处理的各个环节仍然存在很多问题,技术仍然不够成熟,而且随着处理数据量的增加,军事伪装技术的发展以及更高更广泛的处理需求的提出,很有必要研究新的多源图像处理技术。在国家部委预研基金项目的支持下,本文围绕多源图像处理技术展开了一系列理论探讨,主要研究内容包括:1)基于几何特征的多源图像配准算法提出了基于几何特征的多源图像配准算法,并就其中配准点的选取,特征向量的生成、特征配准的实现等进行了详细介绍,并针对配准加速和适用性方面做了讨论。2)图像特征显著度的定义和建模提出了图像特征显著度的概念,用来刻画和度量图像特征在图像处理过程中,对处理结果影响的大小。并以特征点的几何特征显著度为例,通过理论推导完成对图像特征显著度的定义和建模,并给出了完整的建模过程和图像几何特征显著度的计算公式。3)融合图像质量评价指标体系及融合算法评价讨论基于现有融合图像质量评价指标的现状,精选了若干评价指标组成三维度的融合图像质量评价指标体系,包括客观可用维、融合支持维和主观质量评价维。这三个维度从不同的角度刻画融合后图像的质量,这些评价指标相互关联构成一个整体指标体系,可以较全面地对融合图像的质量进行评价。然后简要讨论了图像融合算法的评价指标体系的构建。4)多源图像理论模型和自优化融合框架研究针对目前多源图像融合统一理论模型的研究现状,对现有图像融合的理论模型进行了适当修正,使之适用于多传感器成像的多源图像融合。鉴于多源图像融合的情况复杂性,提出了一种可自动选择合适的多源图像融合算法的自优化多源图像融合框架。该框架通过引入反馈回路,自动对融合规则进行调整和修正,使融合效果达到较好的状态。该框架可以实现多源图像融合过程的自动化和融合状态的自优化,即可自动选择和确定融合算法及融合参数。5)残缺图像特征提取及目标识别技术研究基于相关研究现状,提出了基于拐点的、支持从局部到整体的图像特征提取和目标识别技术。并重点对拐点的确定,拐点特征的提取,拐点特征匹配的距离计算,识别目标的局部吻合度和全局吻合度计算做了详细介绍。本文的研究成果已经在某预研项目中得到了应用,为课题的中期评估和验收打下了重要基础。更多还原

【Abstract】 Multi-source image processing is currently a research hotspot and focus. With the help of the advantage and complementary information of multi-source images, we can generate more complete and more accurate information, thus obtain the more objective and intrinsical understanding of the target in the scene, and reduce the incomplete, uncertainty and even false information which are contained in the spoiled, incomplete, dissemlbling target.So, it has aroused wide concern of scientists from various countries, and has made many achievements. But there are still many problems in the every link of multi-source image fusion; the technology is still not sophisticated enough. Furthermore, with the more and extensive demand are proposed, it is necessary to study new multi-source image processing technology for coping with the increment amount of data processing and the development of military camouflage. In a pre-research foundation project support, centering on the multi-source image processing, this thesis launched a series of theoretical feasibility study, and the main research contents include:1) Multi-source image registration algorithm based on geometric feature of cornersA multi-source image registration based on geometric feature of corners is presented, and the determination of the corners, calculating feature vector of corners and feature matching are described in detail. Finally, the suitability and the acceleration are discussed briefly.2) Mathematical modeling for corner saliencyIn view of the status quo, the definition and model of corner saliency are presented, and the application case is given. The corner saliency can be used to describe and measure the effect degree in image processing results caused by the features. Then taking the geometry feature saliency of corner as an instance, we give the definition of corner saliency firstly, and give the complete modeling process and the calculation formulas of corners saliency late.3) Evaluation criteria architecture of image quality and image fusion algorithm The existing image quality evaluation criteria rarely give consideration to all aspects but just one side simultaneously. Base on the status quo, a 3-dimensions fused image quality evaluation criteria architecture is presented, which is composed of objective usability dimension, fusion support dimension and subjective quality evaluation dimension. These three dimensions can describe the quality of fused image from different angles; they can constitute interrelated evaluation criteria architecture as a whole and evaluate the quality of fused image more comprehensively. Finally, the evaluation criteria architecture of image fusion algorithm is discussed briefly.4) The self optimizing framework of multi-source image fusionAccording to the status quo of multi-source image fusion unified model, an amended unified model was introduced which is suitable for to multi-sensor imaging image fusion. In view of complexity of multi-source image fusion, a self optimizing multi-source fusion framework which can automatically select the appropriate fusion algorithm and parameters was proposed. Because of a feedback loop was introduced in the framework, it can adjust the fusion rules automaticly and amend to the parameter to achieve better fusion effect; it can fuse multi-source images automation.5) Multi-source image feature extraction and object recognitionBased on the related research status quo, this thesis proposed an image feature extraction and object recognition technology which is based on the corner and support“from local to global”to eliminate the impact of military camouflage from occluded images. It places emphasis on the determination of corner, the generation of feature vector, the distance calculating method of corners in feature matching, and the calculation of local goodness of fit and global goodness of fit.Productions of this thesis have been applied in a preliminary research project, and make foundation for the pass of middle examination and final acceptance.更多还原