【作者】 王彦芳；

【导师】 姜威；

【作者基本信息】 山东大学 ， 通信与信息系统， 2011， 硕士

【摘要】 自动聚焦是数字成像系统中必不可少的关键技术之一,近年来得到快速发展,如今已广泛应用在数码照相、显微成像、医学成像、空间探测、器件检测、视频监控、计算机视觉等各个领域。自动聚焦技术的好坏,直接影响着成像的质量和效率,因此很早就成为人们探索的目标和方向。基于图像处理的自动聚焦算法通常分为两类：离焦深度法和聚焦深度法。离焦深度法是一种从离焦图像中获取成像物体深度信息的方法,实现聚焦速度较快,但精度较低；聚焦深度法是一种建立在搜索基础上的聚焦方式,通过评价拍摄图像的清晰度以判断图像是否精确聚焦,从而驱动镜头直至精确聚焦点,精度高,但需要对多幅图像进行质量评价并反馈控制镜头的移动,因此速度较慢。被动式的聚焦深度法,是通过搜寻聚焦评价曲线的峰值实现自动聚焦,因其算法灵活且精度高而被多数自动聚焦系统采用。自动聚焦系统主要包括三个模块：聚焦区域选择、聚焦评价函数和极点搜索算法。聚焦是针对兴趣区域的聚焦,聚焦评价函数完成对兴趣区域的离焦程度评价,极点搜索算法是通过比较评价函数值,反馈控制镜头移动的步长和方向,直至成像质量最佳。通过光学成像分析可知,成像镜头可以等效为低通滤波器,从而得到自动聚焦的原理：图像的高频分量丰富,图像就越清晰；图像的高频分量少,图像就越模糊。因此,聚焦评价函数通常选取高频分量作为聚焦程度的量度。为了提高评价函数尖锐性和抗噪声能力,本文提出了二维加权DCT评价函数,与常用聚焦评价函数进行性能比较和抗噪声能力测试,实验表明二维加权DCT算法和小波变换法具有最佳的尖锐性和抗噪声能力,且相对于小波变换法,二维加权DCT在实时性方面更有优势。由于兴趣区域为前景图像区域,聚焦区域选择算法应选取前景图像区域为聚焦窗口,减少自动聚焦过程中的数据处理量。本文分析了中心取窗、多点取窗、非均匀采样窗口和肤色探测等方式的优缺点,针对这些传统算法的局限性,提出了基于智能优化算法的聚焦区域自适应选择算法；针对标准粒子群算法与人工鱼群算法的自身缺陷,在参数设置、行为方法等方面进行改进,避免陷入局部最优,从而有效地寻找前后景最佳分割阈值,分割得到前景图像,然后选取边缘信息丰富的区域作为聚焦窗口。实验证明,无论成像目标是否位于视场中心,本文算法可以动态地选取目标图像所在区域,具有一定的自适应能力。目前,所采用的聚焦方案一般是单一的聚焦深度法或离焦深度法,基本所有的文献也是在此基础上进行研究,对部分模块算法进行优化。由此,本文提出一种结合离焦深度法速度优势和聚焦深度法精度优势的新的聚焦策略,并在理论上给予推导,初步实验也证明了其可行性。总之,成熟的成像系统和最新的自动聚焦技术基本由日本、美国等一些国家掌握,我国在理论水平和技术水平都还不成熟,所以对自动聚焦的深入研究,既有相当大的理论研究价值,更有广泛的应用价值。更多还原

【Abstract】 Autofocus (AF) is a key technique in digital image capture system and it developed rapidly in recent years. Now the autofocus technique is widely used in many visual applications such as digital cameras, camcorders, video surveillance systems and microscopes. The AF quality directly affects the imaging quality and efficiency, and therefore it attracts us to explore. The AF algorithm based on image processing is usually divided into two classes:depth from defocus (DFD) and depth from focus (DFF). DFD extracts focal depth information from defocus images, and it realizes focus fast but has low precision while DFF is opposite.Because of high precision and flexibly, passive DFF method is used in most autofocus system. An AF system consists of three main modules:a focusing window which defines the region to be focused, a focus measure function (FMF) that evaluates the image sharpness, and a searching strategy to find the global maximum of the FMF. Currently, most imaging systems adopt this semi-digital AF technique, which has the analysis module to determine the focusing status by computing the sharpness of the input image, and the control module to move the lens back and forth until the best focused image is obtained.Imaging lens can be equivalent to a low-pass filter, so the quality of image depends on high coefficients. The FMF usually selects the high coefficients as focus measurement. In order to improve the sharpness and anti-noise ability, this paper proposed a new evaluation function named 2-D weighted DCT. Compared to the common evaluation functions such as gray variance and gradient function,2-D weighted DCT and discrete wavelet transform (DWT) have the best performance. The experiment shows that 2-D weighted DCT performs better in real-time than DWT.Generally, our interest area is the foreground image, so the focus region selection algorithm should adopt the prospects for focus window to reduce data quantity. A variety of window selecting methods have been proposed in the literature such as central window and gauss un-uniformed sampling. For the traditional region-selection algorithm exists some limitations, a method based on intelligence optimization algorithm is proposed; for the standard particle swarm optimization (PSO) and artificial fish swarm algorithm(AFSA) are easy lost in local optimum, the parameters’ setting and action are improved in this paper. Experiment results show that foreground image can be segmented fast and accurately using this method. Focus accuracy of foreground and pixels used in calculating are decreased through the selection of edge region, so the real-time of AF is enhanced; the proposed method can track the main object in image, so the region got in this method is dynamic with good adaptability.At present, adoption of focus scheme is usually single DFF or DFD, and many literatures optimize part module algorithm is based on this study. Thus, this paper puts forward a new focus strategy that combines the high precision of DFF and speed advantage of DFD. Deduction is given in theory and a preliminary experiment also proved its feasibility.Anyhow, the mature imaging system and the latest auto-focusing technologies are controlled by Japan and USA, our country is still not mature in theoretical level and technical level, so thorough research on AF has both theory value and application value.更多还原