【作者】 马书晓；

【导师】 姜威；

【作者基本信息】 山东大学 ， 通信与信息系统， 2013， 硕士

【摘要】 随着时代的进步,我们对各类图像设备(如数码相机、摄像机、影像监视系统和显微镜等)的需求日益增加,自动聚焦技术的研究也变得越来越重要。自动聚焦技术可以使成像设备自动调节直至聚焦以获取清晰的图像,是数字成像过程中的关键技术。自动聚焦技术的发展程度直接影响着图像设备的成像速度和成像质量,因此这项技术的研究具有十分重要的意义。在当前的自动聚焦研究中,聚焦算法主要分为离焦深度法和聚焦深度法两大类。离焦深度法是在离焦图像中直接获取目标物体深度信息的方法,其经过计算得到镜头与目标物体的距离,然后调整镜头到精确聚焦的位置,实现聚焦。聚焦深度法是基于搜索基础的聚焦策略,其通过对多幅图片进行清晰度判断来驱动镜头移动,直至镜头到达精确聚焦点,是当今研究的主流方向,本文也是以聚焦深度法基础的。在本文中,首先对自动聚焦系统的工作原理进行了简单的介绍,数字式自动聚焦系统首先利用感光元件对目标物体进行成像,然后在图像中选择感兴趣区域作为聚焦区域,并用清晰度评价函数对该区域的图像清晰度进行评价,进而根据最大极值搜索法来驱动镜头移动至准确聚焦的位置。自动聚焦系统主要包括三个关键问题：聚焦区域(ROI)的选择,图像清晰度评价函数的选择,和全局最大值搜索算法的选择。在论文接下来的内容中,对这三个问题分别进行了分析。根据对现存的自动聚焦系统关键问题的分析,从聚焦实时性和精确性的角度出发,本论文提出了一种基于提升小波变换的聚焦策略。小波变换是基于有限宽度基函数进行变换的,具有分析频率可调的优点。与傅里叶变换、余弦变换相比,小波变换具有检测图像尖锐边缘处灰度值突变的能力,所以基于小波变换的图像清晰度评价函数可以更好地分辨出图像质量的好坏。提升小波又称第二代小波,其把现存的有限长的小波滤波器分解,加快了小波变换的速度,不再需要二进小波函数平移和伸缩的条件,不依赖于傅里叶变换,但仍具有第一代小波的所有优点,本文中采用提升小波变换进行评价函数的计算。在提出的聚焦策略中,首先利用人工鱼算法选择聚焦窗口,然后在该区域内利用梯度函数作为评价函数,先以自适应大步长穷举搜索算法搜索聚焦点所在区间,然后改用改进后的捉升小波评价函数以小步长爬山搜索算法搜索聚焦点准确位置。在研究过程中采用该聚焦策略对大量的图像序列进行了实验,实验表明,无论目标物体是否在视野中央都能实现精确聚焦,且与传统的小波算法相比聚焦时间可以缩减30%-40%。经初步实验验证,该算法可以避免聚焦点搜索中陷入局部最优的缺陷,具有一定的自适应能力,实时性和准确性也得到了优化。更多还原

【Abstract】 Along with the progress of times, our need to each kind of digital equipments (such as digital camera, DV and image surveillance system and microscope etc.) increases day by day, and the research of auto-focusing also becomes more and more important. Auto-focusing is a key technique in the process of accessing digital pictures, it can make the digital equipments obtain a clear digital picture of the target object automatically. The degree of auto-focusing development directly influence the speed and quality of the digital equipments to get pictures, so the technique of auto-focusing is attached with significance.The current research of auto-focusing is always divide into two kinds:depth from defocus (DFD) and depth from focus (DFF). DFD obtains the focal depth of object target directly from defocus images, it moves the lens to the focused position with the calculation of the distance between lens and object target. DFF is established on the searching pattern, it moves the lens through the information get from the quality of several pictures. DFF is the essential direction of the research, and this text is also based on it.In this text, the principle of auto-focusing system is simply introduced first. Digital auto-focusing system first catches a picture of target object, then select only one small part of the picture as the auto-focusing region and use a focus measure function (FMF) to evaluate the image quality which determine how to move the lens to the focusing position. The auto-focusing system mainly includes three parts:the region of interest (ROI) selection, the focus measure function selection and the maximum searching algorithm selection. The three questions are respectively analyses in the contents behind.According to the analysis of key matters in auto-focus system, the text put forward a new strategy from the point of real-time and precision. Wavelet Transform is based on limited width basis function, it can analyze a signal through adjustable frequency that it can examine the gray changes at the picture edges. Compared with Fourier Transform and Cosine Transform, the Wavelet Transform can examine the picture quality more precisely. Lifting Wavelet Transform (the second generation DWT) divide the existing limited long filter and no longer rely on the Fourier Transform and the condition of translational and flexible. Compared with the first generation DWT. Lifting DWT inherit all the advantages of DWT and it can transform faster than the first generation DWT. According to the characteristics, the text compute the evaluation function through Lifting DWT.The improved auto-focus strategy put forward in the text first selects the ROI with PSO searching algorithm, then uses gradient evaluation function to evaluate the picture quality and searches the focused position with average searching algorithm by big steps. When find the small zone which the focused position lies on, change to use DWT evaluation function to evaluate the picture quality and search the focused position with mountain-climbing searching algorithm by small steps. In the course of the study, many experiments were carried out on the image sequence using the focusing strategy. The results show that, regardless of whether the object is in the center of the field, this strategy can achieve focusing precisely and the focusing time can be reduced30%-40%compared with the traditional wavelet algorithm. According to the experience results, the new auto-focus strategy can avoid the lens find the local focused positon, and it can decrease the time by searching the ROI automatically. The digital picture equipments can achieve auto-focused fastly and accurately with the new strategy.更多还原