【作者】 刘斌；

【导师】 顾越州；

【作者基本信息】 浙江大学 ， 机械电子工程， 2004， 硕士

【摘要】 自动调焦已成为各种成像系统的重要功能，与国外相比，国内在这方面的工作做得还比较少，同时图像处理技术的发展使得自动调焦趋于数字化和智能化，提出基于图像技术的自动调焦方法具有重要的实际意义。 基于图像技术的自动调焦方法采用了与传统调焦技术完全不同的方式进行调焦，传统的调焦方法是通过传感器检测焦点或测量距离的方式实现的，而基于图像技术的调焦方法直接根据图像分析出图像的质量，从而获得当前的成像状态，然后完成调焦操作。图像质量分析是该调焦方法中的关键技术，本文从三种途经详细论述了图像质量分析方法的实现： (1) 基于对比度的图像质量分析方法从图像的时域、频域及信息熵三个角度建立能表示图像对比度的一些调焦函数，并对这些评价函数做了详细的比较，最后确定出时域的Brenner函数和绝对方差函数具有更好的综合性能； (2) 基于功率谱的客观图像质量分析方法假设场景的功率谱具有不变特性，引入了人类视觉系统，加入了维纳噪声滤波器，对图像质量进行评价可得到一个确定的IOM数值，该数值与人的视觉评价具有很高的相关性； (3) 基于小波与神经网络的图像质量分析方法利用小波分析对图像进行多分辨率分解，分析其细节信息并采用统计的方法提取图像特征，再利用人工神经网络对图像特征进行质量模式识别，得到图像的质量等级，实验表明，该方法达到了不错的识别率。 调焦的实现对于对比度法是计算每次成像的调焦函数值，结合一维搜索方法，不断逼近正焦位置；对于功率谱方法和小波与神经网络方法是根据评价出的图像质量，确切地改变焦距调整量。在自动调焦理论的基础上提出了自动调焦系统的设计，分析了系统的总体性能并作为软硬件设计的依据，调焦系统采用DSP+FPGA的高速硬件系统方案。更多还原

【Abstract】 Auto-focus has being an important function in all kind of imaging systems, it has widely application in many fields. In our country, the research and development about auto-focus is poor, but we are facing kinds of application requirement. The image processing is a technology developing rapidly, it promotes the development of auto-focus technology greatly, and numeralization and intelligentization has being a tendence of auto-focus technology. For these conditions, providing an auto-focus means based on image processing technology will have important significance.The auto-focus means based on image technology is different from those traditional auto-focus means entirely. The traditional ones must depend on some special assistant equipments, they uses these equipments to measure distance or find focus and realize auto-focus. However, the auto-focus based on image technology is completed by image quality analysis, it can judge the imaging state from the current image quality, then the system can adjust focus distance correctly supervised by imaging state. From the work flow, we know the image quality analysis is the key technology. In the thesis, three image quality analysis methods are presented detailedly:(1) The first one is the image quality analysis based on image contrast, it analyzes kinds of functions working at temporal field, frequency field and information entropy, compares their performance in the number of calculation and the time consumption of calculation, and finds out two better functions.(2) The next one is the image quality analysis based on image power spectrum, it supposes nature scenes have the same power spectrum, embeds a human vision system and includes a Winner de-noise filter. It analyzes a picture and gets an IQM (Image Quality Measure) value, the IQM value is highly relative to the estimation result through human vision system.(3) The last one is the image quality analysis based on wavelet analysis and neural network, it analyzes image from different frequency resolution through wavelet theory, and abstract the image characteristics from these detail information through statistics analysis, then uses an artificial neural network for quality pattern recognition. The recognition rate of this means is good.The realization of auto-focus, for contrast means, is similar with the 1-demension search, calculates the auto-focus function value, compares it with the previous one and decides the next position; for power spectrum or wavelet means, moves a special distance relative to the image quality level directly.Depending on above auto-focus theories, design an auto-focus system, analyze its performance and present the design of hardware and software particularly, the system is constructed by high speed DSP (Digital Signal Processor) and FPGA (Field Programmable Gate Array).更多还原