声纳图像配准与融合技术研究

发布时间：2018-05-28 20:25

本文选题：图像配准 + 互信息　；参考：《哈尔滨工程大学》2014年博士论文

【摘要】：声纳图像直观地表征了水下声探测中的目标回波数据,因此对其进行配准与融合在后期的图像处理和分析中起着关键作用。由于声成像的原理导致声纳图像具有一些特有的属性,由此造成声纳图像配准与融合的困难,同时也对处理过程提出了一些要求或前提条件。虽然在光学,遥感,医学等其他领域已经涌现出大批切实的图像配准与融合方法,然而探究关于声纳图像的处理策略和方式依然具有广阔的应用前景和迫切的现实需求,因此,本文针对声纳图像配准融合问题展开了有意义的研究。在声纳图像配准方面,本文选取图像配准领域中通行的互信息理论作为基础。针对刚性形变图像配准,主要探究了相似度量函数和优化方法两方面内容,一方面是兼顾了像素点空间联系的度量函数,另一方面是避免了度量函数陷入局部极值的优化策略。这些方法中,不乏有非常成熟或已提出多年的模型,但在声纳图像配准中却鲜有应用。本文针对声纳图像配准问题,分别探讨了各式配准方法的性能,进行了大量的仿真实验,并在基于互信息的图像配准框架下,推导了基于梯度互信息的相似度量函数的具体表达式以及建立了基于粒子群算法和Powell法的混合优化策略,利用多分辨率分析方法改善声纳图像配准算法的性能。在归纳出声纳图像配准方法后,对具有代表性的声纳图像进行配准实验,在刚性形变声纳图像配准中,通过对比其他配准算法,本文算法表现性能优良且鲁棒稳定。此外,在刚性形变声纳图像配准基础上,考虑对非刚性形变的声纳图像进行配准。分析了 Demons算法在声纳图像配准中的局限和不足,构造了基于梯度互信息的Demons模型,并提出了相应的声纳图像配准算法,同时本文结合声纳图像特点,对Demons算法的重要系数进行了分析,得出了有价值的结论。采用基于梯度互信息的Demons模型对非刚性形变的声纳图像进行配准,通过与经典和主动Demons模型进行对照,验证了其有效实用。在声纳图像融合方面,本文从融合和评价两个方向进行探索。起初探讨了以Piella理论为基础的通用融合框架在声纳图像融合中的应用,全面系统地阐述了基于Piella多尺度图像融合框架,包括多分辨率分析方法,活性测度,匹配测度,决策模块以及合成模块。针对声纳图像及其处理过程的自身特点,本文扩展了 Piella多尺度图像融合框架,摒弃匹配测度和决策模块,并提出引入前一分解层中的近似信息构造低频分量的融合准则以及基于图像质量评价标准的高频融合准则。仿真实验验证,这些改进对声纳图像融合具有重要意义,改进后的算法适于声纳图像融合处理,更有利于融合质量的提高,取得了良好的融合效果。其次,本文系统地研究和分析了各种图像融合评价标准的主客观方法,并总结了适合声纳图像融合评价的一般性原则。基于此原则,对多种经典图像融合方法进行了深入细致地分析,并得出相应的结论。最后,提出了一种基于区域相似性的图像融合评价方法,经仿真实验验证其具有与人类视觉系统良好的一致性。
[Abstract]:Sonar image is intuitionistic on the target echo data in underwater acoustic detection, so registration and fusion play a key role in the later image processing and analysis. The acoustic imaging principle causes the sonar image to have some unique attributes, which causes the difficulty of sonar image registration and fusion, and also has been dealt with. Although a large number of practical image registration and fusion methods have emerged in other fields, such as optics, remote sensing, medicine and other fields, the study of sonar image processing strategies and methods still has broad application prospects and urgent practical needs. Therefore, this paper aims at sonar image registration and fusion. In the aspect of sonar image registration, this paper selects the mutual information theory in the domain of image registration as the basis. Aiming at the rigid deformed image registration, it mainly explores the two aspects of the similarity measure function and the optimization method, on the one hand, it takes into account the metric function of the spatial connection of pixels, on the other hand. In these methods, there are many models which have been very mature or have been proposed for many years, but they are rarely applied in sonar image registration. In this paper, the performance of various registration methods is discussed for sonar image registration problem, and a large number of simulation experiments are carried out, and the mutual information is based on mutual information. In the frame of image registration, the specific expression of the similarity measure function based on the gradient mutual information is derived, and a hybrid optimization strategy based on particle swarm optimization and Powell is established, and the performance of the sonar image registration algorithm is improved by using the multi-resolution analysis method. On the basis of rigid deformed sonar image registration, the registration of sonar images with non rigid deformation is considered on the basis of rigid deformed sonar image registration. The limitations of Demons algorithm in sonar image registration are analyzed. The Demons model based on the gradient mutual information is constructed and the corresponding sonar image registration algorithm is proposed. At the same time, this paper analyzes the important coefficients of the Demons algorithm by combining the features of sonar image, and draws a valuable conclusion. The Demons model based on the gradient mutual information is used to match the sonar images with non rigid deformation. By comparing with the classic and active Demons models, it is proved that it is effective and practical. In the aspect of sonar image fusion, this paper explores two directions from fusion and evaluation. At first, the application of general fusion framework based on Piella theory in sonar image fusion is discussed, and the multi scale graph based on Piella is systematically expounded. Image fusion framework, including multi-resolution analysis method, activity measure, matching measure, decision module and synthetic module. Aiming at the characteristics of sonar image and its processing, this paper extends the Piella multi-scale image fusion framework, abandons the matching measure and decision module, and proposes the introduction of low approximate information in the former decomposition layer. The fusion criterion of frequency component and the high frequency fusion criterion based on image quality evaluation standard. Simulation experiments show that these improvements are of great significance to sonar image fusion. The improved algorithm is suitable for sonar image fusion, which is more conducive to the improvement of fusion quality and good fusion effect. Secondly, this paper systematically studies and studies the fusion quality. This paper analyzes the subjective and objective methods of various image fusion evaluation criteria, and summarizes the general principles suitable for the fusion evaluation of sonar images. Based on this principle, a detailed analysis of various classical image fusion methods is carried out and the corresponding conclusions are drawn. Finally, a method of image fusion evaluation based on Regional similarity is proposed. Simulation experiments show that it has good consistency with the human visual system.
【学位授予单位】：哈尔滨工程大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP391.41;TB56

【相似文献】