【作者】 王晓玲；

【导师】 李庆武；

【作者基本信息】 河海大学 ， 通信与信息系统， 2007， 硕士

【摘要】 小波分析理论由于其卓越的时频分析特性，在数字图像处理领域中倍受青睐。关于小波分析理论解决实际问题的研究具有重要的理论意义和应用价值。本文涉及的课题是用小波分析的方法对复杂环境下车辆图像进行研究，实现了图像采集、图像质量评估、图像处理、车型识别四个功能。 在Windows系统平台上，通过人机界面的方式利用Visual C++6.0集成开发工具对图像采集卡进行二次开发，编写包含设备设置、图像格式设置、字符叠加、图像压缩以及图像保存等功能的图像采集模块的软件，给出了图像采集系统界面。 对采集到的实际车辆图像进行无参考图像质量评估。根据图像信号和噪声在小波域的特性，研究出一种评估图像噪声强度的方法。根据模糊图像的频谱和协方差特性估计模糊图像的模糊度。最后用灰度直方图来评估图像对比度。 根据图像质量评估的结果对质量低的车辆图像进行图像处理。研究了改进的指数阈值降噪算法和平移不变量小波阈值降噪算法，对于含噪声的车辆图像同时作用两种算法较已有的降噪算法能够有效的抑制噪声、保留细节。对于模糊车辆图像，根据模糊度的参数构造引起图像退化的点扩散函数，然后进行维纳滤波，从而去除图像模糊。对于对比度低的车辆图像，使用基于小波分析的图像增强算法，得到很好的视觉效果。 最后实现货车、客车、轿车三种车型的识别。在识别图像中提取车辆，用小波边缘检测法检测到车辆边缘，对边缘图像进行膨胀和填充获取车辆轮廓，从而提取车辆特征，最后将特征向量输入LVO神经网络对车辆进行车型识别。更多还原

【Abstract】 Wavelet analysis theory is favored in digital image processing domain because of its excellent Time-Frequency characteristic. It has great theoretical significance and practical value to solve practical problems using wavelet theory and digital image processing technology. This dissertation mainly achieves four functions: image capture, image quality assessment, image processing and vehicle type recognition.Based on Windows operating system, the first task is the secondary development of image capture card through VC++6.0, MV-4000 Capture Card is used. The program module includes: equipment set, image format set, save image and so on.Three no-reference image quality assessment criterions for captured vehicle image are proposed in this dissertation. Because of the characteristic of signal and noise in wavelet domain, a quality assessment method for noised image is achieved. According to the frequency spectrum and covariance, the blur strength of the blurred image is evaluated. Then contrast assessment method by means of gray histogram is presented.The degraded images are processed according to the result of image quality assessment. This dissertation studies improved exponential threshold de-noising method and TI (Translation invariant) wavelet de-noising method, the methods not only can restrain noise, but also can reserve details. For the blurred image, structure PSF (Point Spread Function) according to the blur factors, then de-blur the image using Wiener Filter. A wavelet-based contrast enhancement method is researched. The result demonstrates the algorithm can avoid over-enhancement of noise while enhancing contrast of the image with better visual effect.At last, the function of vehicle type recognition is realized. The vehicle is separated at first, and the vehicle edge is extracted by B-spline wavelet edge-detection method, and the vehicle edge is dilated and filled, and the outline is achieved, and the features are extracted. Finally, feature vector is imported to LVQ neural network, and three types which are truck, car, bus, can be recognized.更多还原