基于图像处理技术的自动调焦方法研究

发布时间：2018-09-04 16:08

【摘要】：图像是信息传递的重要载体,如何获取清晰、高质量的图像画面一直是人们探索的方向与目标,图像质量直接关系着后续的图像处理与应用。自动调焦技术作为光学成像系统的关键技术之一,广泛应用于医学显微镜、扫描仪等精密仪器,以及卫星导航、空间探测、机器人视觉、自动监控等领域。基于图像处理的自动调焦技术不同于传统调焦技术,该技术仅依赖于获取的图像信息,不需要额外的光电检测设备,按照图像信息处理方式可分为两大类:离焦深度法和聚焦深度法。本文主要就聚焦深度法进行了深入研究。基于聚焦深度法的调焦过程是一个闭环控制过程,集光、机、电于一体,主要包括三个重要环节:调焦窗口的构建、调焦评价函数的选取、搜索算法对调焦电机的反馈控制。本文在研究国内外自动调焦技术发展现状基础上,针对这三个重要环节,并结合实际应用开展了一系列研究。首先,建立了光学成像系统的离焦模型,介绍了离焦状态下的点扩散函数与光学传递函数,通过分析离焦造成图像模糊的本质原因,奠定了自动调焦技术的理论基础。其次,对调焦评价函数进行了深入研究。调焦评价算法的选取是调焦过程中一个至关重要的步骤。离焦会造成图像高频分量的流失,在空间域上表现为图像边缘及细节变模糊、图像质量下降。本文分析了常见的基于图像梯度、频率域、信息熵、统计学的调焦评价函数。基于傅里叶变换的图像功率谱具有场景不变性的特点,有着广泛应用。小波分析克服了傅里叶单一分辨率的缺陷,具有多分辨率分析的特点。小波功率谱函数是对基于傅里叶变换的功率谱函数的改进。结合人类视觉系统(HVS)的低通滤波特性,对小波功率谱法进行加权处理。使得评价结果更加符合人眼视觉感知特性。为了验证算法的有效性及通用性,在LIVE数据库上进行了相关实验。同时,针对大型光测设备的特点,本文又提出一种基于自适应阈值分割的改进SML清晰度评价函数,在原有的SML函数基础上增加两个对角方向的梯度值,并取最大值作为该点空间滤波的结果,该算法能有效地克服背景噪声、以及大气湍流造成设备抖动的影响,具有较高的稳定性与灵敏度。再次,对调焦窗口的构建进行了研究。讨论了调焦窗口的尺寸及位置对调焦评价曲线的影响。由于光测设备的观察目标具有位置及尺寸上的不确定性,窗口过大或过小容易造成背景信息的干扰或目标缺失。在分析比较传统调焦窗口选择方法局限性的基础上,提出了基于多尺度脉冲余弦变换的窗口构建方法。该方法模拟人类视觉注意机制,从而提取图像感兴趣区域,能使光测设备在任意调焦状态下很好地适应成像主体的位置变化,建立有效调焦窗口。然后,制定了自动调焦系统的搜索策略。搜索算法是调焦过程的最终实现。本文针对比较常用的、具有较高可行性的爬山法进行了改进。以改进爬山法作为调焦系统的基础搜索算法,同时以图像的小波功率谱值作为辅助手段,综合判断调焦搜索方向以及极值点性质;并采用自适应步长,提高了搜索速率。从而快速准确的得到最佳聚焦位置,避免了范围较大的局部极值点的干扰。最后,本文根据所提出的调焦策略,设计了以DSP芯片TMS320C6678和Virtex-II系列XC2V3000为核心的嵌入式自动调焦系统,并在调焦精度、稳定性、实时性等方面进行了大量的测试实验。实验结果表明本文所提出的自动调焦方法切实可行,具有较高工程应用价值。
[Abstract]:Image is an important carrier of information transmission. How to obtain clear and high-quality images has always been the direction and goal of people's exploration. Image quality is directly related to the subsequent image processing and application. Auto-focusing technology based on image processing is different from traditional focusing technology, which only relies on the acquired image information and does not require additional photoelectric detection equipment. Focusing process based on depth of focus method is a closed-loop control process, which integrates light, machine and electricity. It mainly includes three important links: the construction of focusing window, the selection of focusing evaluation function, and the feedback control of search algorithm for focusing motor. Based on the development status of focusing technology, a series of researches are carried out aiming at these three important links and combining with practical application. Firstly, the defocusing model of optical imaging system is established, the point spread function and optical transfer function under defocusing state are introduced, and the auto-focusing technology is established by analyzing the essential reasons of blurring image caused by defocusing. Secondly, focusing evaluation function is studied deeply. The selection of focusing evaluation algorithm is an important step in the focusing process. Defocusing will cause the loss of high-frequency components of the image, which will blur the edges and details of the image and degrade the image quality in the spatial domain. Fourier transform based image power spectrum has the characteristics of scene invariance and is widely used. Wavelet analysis overcomes the shortcomings of single Fourier resolution and has the characteristics of multi-resolution analysis. Wavelet power spectrum function is a power spectrum function based on Fourier transform. In order to verify the validity and universality of the proposed method, some experiments are carried out on LIVE database. At the same time, according to the characteristics of large-scale optical measurement equipment, a new method is proposed in this paper. The improved SML intelligibility evaluation function based on adaptive threshold segmentation adds two diagonal gradient values to the original SML function and takes the maximum value as the result of the point space filtering. The algorithm can effectively overcome the background noise and the influence of equipment jitter caused by atmospheric turbulence, and has high stability and sensitivity. Thirdly, the construction of the focusing window is studied. The influence of the size and position of the focusing window on the focusing evaluation curve is discussed. Due to the uncertainty of the position and size of the object observed by the optical measuring equipment, too large or too small of the window can easily cause the interference of background information or the loss of the object. Based on the limitation of the method, a window building method based on multi-scale pulse cosine transform is proposed. This method simulates the human visual attention mechanism and extracts the region of interest (ROI) of the image. The search strategy of the dynamic focusing system is the final realization of the focusing process.This paper improves the hill-climbing method which is commonly used and has high feasibility.The improved hill-climbing method is used as the basic search algorithm of the focusing system,and the wavelet power spectrum value of the image is used as the auxiliary means to judge the direction of the focusing search. Finally, according to the proposed focusing strategy, an embedded auto-focusing system based on DSP chip TMS320C6678 and Virtex-II series XC2V3000 is designed. A large number of experiments have been carried out on focusing accuracy, stability and real-time. The experimental results show that the auto-focusing method proposed in this paper is feasible and has high engineering application value.
【学位授予单位】：中国科学院研究生院(长春光学精密机械与物理研究所)
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】