基于GPU的数字全息自动聚焦技术研究

发布时间：2018-08-02 10:58

【摘要】：数字全息具有快速、无接触、全场、三维实时测量等许多优点,广泛应用于微小器件检测、粒子场分析、生物显微观察、光学器件检测等方面。虽然数字全息可通过计算机完成待测物体再现,但为了获得最佳再现像质量,需通过自动聚焦技术寻求最佳再现距离。对于数字全息自动聚焦而言,需重点考虑两个指标:一是能否准确找到最佳再现距离,二是能否快速完成聚焦。课题“数字全息自动聚焦技术研究”在研究自动聚焦涉及的全息图记录和再现、聚焦评价函数和聚焦搜索算法等基础上,改进了一种基于db4小波基的聚焦评价函数,在准确判断最佳再现距离的同时,获得强的抗噪性;进而利用GPU的高性能并行快速计算特点,实现自动聚焦算法优化加速。论文的主要内容如下:首先,在分析数字全息图记录和再现的原理基础上,通过两种数字全息记录结构和三种再现算法的对比与分析,确定了离轴全息光路结构和菲涅耳变换法为课题研究所使用的结构和算法;详细阐述了常用的灰度方差法、梯度平方法、傅里叶频谱对数法等聚焦评价函数的原理及其定义形式。其次,在分析自动聚焦方案涉及的聚焦评价函数和聚焦搜索算法两部分关键技术基础上,依据小波变换对图像时频局部化分析的优点,提出了一种基于小波变换的聚焦评价函数,并与其它传统聚焦评价函数对比验证了其性能的优越,能够准确找到全息图的最佳再现距离;在给出遍历搜索法、Fibonacci搜索法和“盲人”爬山搜索法三种常用的搜索算法原理基础上,分析了其优缺点,并给出其适用范围。最后,将自动聚焦算法移植到GPU上,利用GPU强大的并行计算能力对算法加速,与在CPU上运行相比,自动聚焦算法的运行速度得到提升了 10倍以上,实现了快速聚焦的目标。以GPU的自动聚焦程序原型开发了一款数字全息自动聚焦应用软件,软件界面简单易懂,方便用户使用。
[Abstract]:Digital holography has many advantages, such as fast, contactless, full-field, three-dimensional real-time measurement and so on. It is widely used in micro-device detection, particle field analysis, biological microscopic observation, optical device detection and so on. Although digital holography can reproduce the object to be measured by computer, in order to obtain the best image quality, the best reconstruction distance must be obtained by autofocus technique. For digital holographic auto-focusing, two indexes should be considered: one is whether the best reappearance distance can be found accurately, and the other is whether the focus can be completed quickly. On the basis of studying the recording and reproducing of holograms, focusing evaluation function and focusing search algorithm, a focus evaluation function based on db4 wavelet basis is improved. At the same time, the best reappearance distance is judged accurately, and the strong noise resistance is obtained, and then the optimization and acceleration of the autofocus algorithm is realized by using the high performance parallel and fast computing characteristics of GPU. The main contents of this paper are as follows: firstly, on the basis of analyzing the principle of digital hologram recording and reproducing, the comparison and analysis of two kinds of digital holographic recording structures and three reproducing algorithms are made. The structure and algorithm of off-axis holographic optical path structure and Fresnel transform method are determined, and the commonly used grayscale variance method and gradient leveling method are described in detail. Secondly, on the basis of analyzing the two key technologies of focusing evaluation function and focusing search algorithm, the advantages of wavelet transform in image time-frequency localization analysis are analyzed. A focus evaluation function based on wavelet transform is proposed, and compared with other traditional focus evaluation functions, its performance is proved to be superior, and the best reconstruction distance of hologram can be found accurately. On the basis of three common search algorithms, the traversal search method, Fibonacci search method and the "blind" mountain climbing search method, the advantages and disadvantages of these algorithms are analyzed, and the applicable range is given. Finally, the autofocus algorithm is transplanted to GPU, and the algorithm is accelerated by the powerful parallel computing power of GPU. Compared with running on CPU, the speed of autofocus algorithm is increased by more than 10 times, and the goal of fast focusing is achieved. A digital holographic auto-focusing application software is developed based on the autofocus program prototype of GPU. The software interface is easy to understand and easy to use.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;O438.1

【参考文献】