稀疏表述理论在关联成像中的应用
文献类型：学位论文
| 作者 | 徐旭阳 |
| 学位类别 | 博士 |
| 答辩日期 | 2015 |
| 授予单位 | 中国科学院上海光学精密机械研究所 |
| 导师 | 韩申生 |
| 关键词 | 关联成像 稀疏表示 字典 相干度 |
| 其他题名 | The application of sparse representation theory in ghost imaging |
| 中文摘要 | 关联成像，也称为鬼成像，是一种利用光场涨落的二阶强度关联得到 物体图像的非局域成像技术。自上世纪 90年代以来，由于其独特的成像结 构以及机理，在遥感成像、医学成像、X射线成像等方面有着广泛的应用前 景。但是基于关联算法的传统关联成像存在成像质量差、所需的采样数目 比较多等弊端，因此可以将图像的稀疏先验特性引入到关联成像中，这样 可以有效的减少采样的数目，提高目标图像的信噪比，而且可以实现对目 标特定结构的成像。论文的内容安排如下所示： 第1章，由于关联成像在遥感成像、医学成像、X射线成像等领域有着 比较广泛的应用前景，近年来，关于关联成像的研究越来越被研究人员重 视，但是，传统的关联成像有着成像质量差，采样效率低的缺点，因此对 于如何提高其采样效率成了一个研究热点，本章中首先讨论了传统的关联 成像的性质，成像机理以及国内外的发展现状，然后介绍基于稀疏限制下 的关联成像的原理、算法等，将图像的稀疏特性融入关联成像之后可以有 效的提高成像的采样效率以及质量。 第 2 章，将图像的稀疏特性引入关联成像的过程通常涉及到对图像的 各种数学变换，利用这些变换可以将图像用比较少的系数尽可能的描述出 来，这些变换又可称为字典，因此本章中首先介绍各种数学变换如离散余 弦变换、小波变换、曲线波变换等，然后再讨论这些变换对于关联成像恢 复结果的影响。由于实际中的图像的像素比较大，如果写出这些变换的显稀疏表述理论在关联成像中的应用/徐旭阳 II 示表达形式，那么其计算量以及存储量非常大，所以需要找出一种比较快 速的算法，使我们在计算的过程中能够较快的实现这些变换，节约计算成 本。 第 3 章，在实际的成像过程中，有时我们并不需要目标区域的所有图 像信息，需要的只是局部区域内的一些信息，甚至其它的信息可以认为是 一种噪音或者干扰，那么在这种情况下，我们需要在关联成像的过程中对 目标的各种不同特征实现分离，在本章中，将形态分量分离的方法引入基 于稀疏限制的关联成像中，这样可以实现对于目标区域的特征分离，而且 将其不同分量叠加之后还可以达到一定的去噪效果，可以为后期利用关联 成像实现目标的识别与跟踪做铺垫。 第 4 章，在传统的关联成像中，成像系统都是双臂的，一条光路探测 物体表面的散斑的光强分布，另一条光路探测光场经过物体或由物体反射 之后的光强，随着关联成像的发展，由于需要考虑到物体的稀疏特性，因 此在图像的恢复过程中引入了稀疏矩阵，在这种情况下，我们可以通过固 定稀疏矩阵，然后优化该稀疏矩阵与探测矩阵的相干度来提高系统的采样 效率以及恢复图像的信噪比，利用优化后的探测矩阵来设计一种光学调制 器件，通过该器件来控制物体表面散斑的光强分布情况，这样可以提高我 们恢复的图像的信噪比. 第 5章，论文的总结以及对关联成像的前景展望 |
| 英文摘要 | The correlation imaging which is also termed as ghost imaging is to imaging by the second intensity correlation of fluctuation of light field. It’s a kind of imaging technology unlocally. As a result of specially structure and imaging mechanism, it has many potential applications in the area of remote sensing, biomedical imaging, X-ray imaging since 90’s in last century. The traditional ghost imaging has many weakness, such as poor reconstructed image, low sampling efficiency and so on. Then the sparsity of nature image has been introduced into the ghost imaging. Researchers have found that high sampling efficiency and better quality of reconstructed image can be realized in this situation. It can also be imaging with the special part of objet. The content of thesis is as follows: Chapter 1. Since the potential applications in the area of remote sensing, biomedical imaging and X-ray imaging, more and more attention has been paid on the ghost imaging. The reconstructed image of traditional ghost imaging has bad quality and low efficiency. Technology of increasing the quality of image has become a research hotspot now. In the chapter, the property, imaging mechanism and background have been introduced first, then are the theory and reconstruction algorithm of ghost imaging via sparsity constraint. The quality of reconstructed image and sampling efficiency can increase obviously after the combination of sparsity of image with ghost imaging. Chapter 2. The introduction of sparsity is connected with many mathematical transformation of image. The image can be descripted by as fewer as coefficients in the sparsity representation. The transformation can also be termed as dictionary. In this chapter, the dictionaries such as discrete cosine transform (DCT), discrete wavelets transform (DWT) and curvelets transform have been introduced and their effects on the ghost imaging have been studied. It’s difficult to store and computer the representation if the size of image is large. So the fast algorithms for these dictionaries are need in our reconstruction. Chapter 3. In the actually situation, we don’t need the all information of the whole field of view. The local part of image is needed for some purpose, other part can be thought as an interference or noise. Under the situation, the separation of different parts is need in our ghost imaging. In this chapter, the morphology component separation has been introduced into the ghost imaging. The separation of different parts has been realized in ghost imaging. It’s helpful for the target recognition and tracking of ghost imaging later on. Chapter 4. There are two arms in ghost imaging. One area-array detector detects the speckle of light transmitted some distance, the other one detects the light intensity reflected or transmitted by the object. With the develop of ghost imaging, the sparsity of image has been taken into consideration during the reconstruction. In this situation, the dictionary need to be chosen before the sampling according the sparsity prior of image. The sensing matrix can be optimized before sampling according its mutual coherence with dictionary. In this chapter, the mutual coherence of sensing matrix and dictionary has been decreased by the gradient descent algorithm. The reconstruction with better quality of image and higher efficiency can be realized by the optimization sensing matrix. This matrix can be used to fabricate a kind of optical modulation component to control the light speckle in ghost imaging. This component can help to get a better quality of image in our ghost imaging via sparsity constraints. Chapter 5. The conclusions of the thesis and future prospects are given in this chapter. |
| 语种 | 中文 |
| 源URL | [http://ir.siom.ac.cn/handle/181231/15931]  |
| 专题 | 上海光学精密机械研究所_学位论文 |
| 推荐引用方式 GB/T 7714 | 徐旭阳. 稀疏表述理论在关联成像中的应用[D]. 中国科学院上海光学精密机械研究所. 2015. |
入库方式： OAI收割
来源：上海光学精密机械研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。