基于数字图像处理的自动调焦技术研究

	基于数字图像处理的自动调焦技术研究
	刘雪超
学位类型	博士
导师	吴志勇
	2014-07
学位授予单位	中国科学院大学
学位专业	电路与系统
摘要	图像在信息化社会中起到重要的作用，尤其是对于军事领域的大型光测设备，清晰的图像画面可为指挥控制中心提供最直接、最有价值的决策依据。然而，根据成像的原理，任何光学系统在成像过程中都不可避免地遇到离焦的问题。成像系统离焦会使采集到的图像呈现模糊失真，影响后续的传输和处理，尤其在战时甚至会影响到指挥决策的做出。因此，为克服这一弊端，应在光测设备执行任务的同时进行实时的自动调焦，避免因离焦失真带来的影响。传统的调焦方法往往通过测距或相位检测实现，提高对设备的硬件要求，增加成本。本文结合大型光测设备的实际应用，从图像处理的角度出发，采用聚焦深度法，以采集到的图像为依据，在不增加复杂硬件和辅助工具的前提下，实现自动调焦，提高光测设备的智能化。根据光学成像系统的原理将基于图像的自动调焦过程分为四个模块：图像预处理、调焦窗口构建、图像质量评价和正焦位置搜索。通过研究国内外自动调焦的发展历程，对比各方法的优劣，结合实际工程任务要求对各模块进行研究，并提出相关算法及整体方案。图像预处理的目的是克服噪声和曝光不当对光测设备成像的影响，提高图像质量和识别度。本文针对光照干扰提出具有动态窗口的二维直方图解决方法，使图像中的对比度得到合理增强，校正了曝光不当对成像效果的影响。调焦窗口的构建使后续调焦过程仅针对目标进行，有助于降低运算量，克服背景和噪声的影响。针对光测设备所观测目标的位置和大小具有随机性的特点，本文从生物视觉角度出发，提出基于视觉感知的调焦窗口构建方法，以Itti模型为基础模拟人眼观察过程，提取图像信息特征，并通过去噪和模板扩展法建立规则区域，实现对任意调焦状态下调焦窗口的构建。同时，本文又从分类的角度出发，提出结合支持向量机制(SVM)的调焦窗口构建方法，再以分支定界的搜索原则提高目标位置检测的效率。图像质量评价是调焦过程的关键步骤，评价效果的好坏直接影响到整个调焦过程成功与否。本文提出结合NSS和空域变换的无参质量评价方法，在空域变换中提取自然图像的统计特征，再与参考特征相比较，实现无参质量评价，克服图像内容的影响，使光测设备在调焦初始阶段仅需一帧图像就能确定离焦程度，为搜索策略的制定提供有效依据，避免了盲目性。同时，为降低耗时和保障调焦精度，本文又提出改进的自相关质量评价方法，扩大函数的原有尺度，增加了垂直和对角方向的相关性，提高评价函数的稳定性和对清晰度的灵敏度，克服调焦过程中光测设备受大气湍流和自身抖动的影响。搜索是整个调焦的最终实现，搜索策略的制定应结合实际设备和观测情况。本文的搜索策略采用粗调焦和精调焦相结合的方式，以爬山搜索法为基础，优化搜索步长，在粗调焦中采用结合NSS和空域变换的方法进行质量评价，得到离焦程度和调焦方向；在精调焦阶段采用改进的自相关函数进行评价，降低耗时、保障精度。本文最后结合光测设备的实际应用提出基于图像处理的自动调焦整体方案，设计了硬件系统结构和软件流程，并进行了大量的实验验证，分析了调焦效果、精度、稳定性和实时性，建立了统计性数据。结果显示，本方法切实可行，具有较高的实际工程应用价值。
其他摘要	Image is playing very important part in this information-based society, especially for the equipment in military field. Clear image could provide useful information directly for decision-making by headquarters. Actually, there is no kind of optical system can avoid defocus problem because of the imaging principle. The image is blur distortion if the optical measurement system is in defocus state. It could exert an influence on subsequent processing and transition, especially effect the decision-making during war time. Consequently, real-time auto-focus is essential for optical measure systems performance to avoid disadvantage of blur distortion. Traditional method always solves this problem by distance measurement or phase detection, which needs extra hardware and increases costs. This paper studies the project from image processing based on optical measuring equipment. By depth from focus, auto-focus can be realized just based on image information without any other auxiliary tools, which makes the optical measure systems more intelligent. Based on imaging principle, the auto-focus procedure includes four modules: pre-processing, focus window construction, image quality assessment and searching for focal position. This paper proposes solution to every module and the overall program around this topic with the practical engineering requirements after studying development history of this field at home and abroad and comparing advantage and disadvantage of many methods. Pre-processing is needed to overcome influence of noise and inappropriate exposure during optical measure systems imaging, meanwhile, can enhance image quality. To solve the inappropriate exposure, a novel two-dimensional histogram method with dynamic window is proposed. This enhances contrast and correct exposure. Focus window construction is useful for refraining from affecting of noise and background. At the same time, this could reduce the amount of calculation and makes the following steps just for target region. There is a feature that the position and size of target is random duiring measurement. This paper presents a novel method from biological vision to construct focus window for the feature. The method simulates visual perception based on Itti model, extracts feature message and constructs regular window via extension by cross-shaped template. It can construct focus window for any focus state. Meanwhile, this paper provide a way by SVM from classification to build focus window and invites branch and bound method to optimize searching process, which can improve location efficiency for target. Image quality assessment（IQA）is an critical step for focusing. And the IQA result can affect the focusing result. This paper proposes a novel method to get no-reference IQA based on natural scene statistics（NSS）and spatial transition. This can overcome the effect of content and decide the defocus extent just by only one frame in primary stage of focusing for optical measurement. Then, it can direct searching process and avoid blind seeking. Then, the paper improves self-correlation function for IQA by larger scale and more orientation pertinence. It can make the IQA result stable and be sensitive to sharpness, meanwhile, overcome the influence of atmospheric turbulence and equipment jitter to optical measurement. Searching is the final step for focusing. Search strategy should be made with consideration of equipment and application. This paper adopts rough focusing and fine focusing based on hill climbing search method to optimize searching step length. In rough process, the method of combining NSS and spatial transition is applied for IQA. In fine process, the method of advanced correlation function is applied for IQA. The first one is used to defocus degree and direction. The second one is used to reduce costs and ensure accuracy. Finally, this paper provides the whole auto-focus program based on image processing for optical measurment. The program includes hardware and software structure. After a lot of experiments, the paper analyses focusing results, accuracy, stability and real time feature, meanwhile, makes statistic dates. The results show that the program is feasible and useful for engineering.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/41442
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	刘雪超. 基于数字图像处理的自动调焦技术研究[D]. 中国科学院大学,2014.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
刘雪超.pdf（4125KB）			开放获取	CC BY-NC-SA	请求全文