【作者】 张敏；

【导师】 蒋建国； 齐美彬；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2012， 硕士

【摘要】 随着科技的发展，人们对图像质量的要求也越来越高。在特定的应用场合，图像采集设备得到的图像质量比较低，造成细节信息丢失，加上光照的干扰，使得图像对比度降低。我们需要从退化图像中，得到细节信息丰富的图像。交通、医学图像领域，更关注边缘和细节信息的提取，期望获得违规车辆的车牌号，或人体病变位置的信息，因此对降质图像进行复原，就十分必要了。图像复原旨在由降质图像重建清晰图像，当被拍摄物体与相机之间存在相对运动时，就会产生运动模糊，导致图像质量下降，由于运动模糊的普遍存在性，通过算法对图像进行去模糊，以期得到清晰的图像。论文分析了运动模糊产生的原因，建立其退化模型，对该模型中的点扩散函数进行估计，具体表现为模糊角度和模糊尺度估计，最后采用维纳滤波复原，完成整个图像去模糊过程。针对图像复原中产生的振铃效应，分析其产生的原因，在获取、传输、存储图像过程中，不可避免地引入噪声，因此，在图像预处理阶段通过haar小波去噪，以抑制振铃效应的产生。论文完成了对图像的去噪和去模糊，具体工作如下：1)采用Gabor滤波器估计运动模糊的角度参数。先将图像频谱图与16个Gabor滤波器进行卷积，再根据最大响应对应的滤波器方向得到每点的角度方向，去除弱方向点后，将各点角度平均值作为模糊角度。2)采用自相关函数计算运动模糊尺度。先对图像进行水平、垂直方向差分运算得到差分图像，再计算得到每行差分图像的自相关函数，并将全部自相关函数相加得到累加自相关函数曲线，根据该曲线的对称负峰值之间的距离计算模糊尺度。实验结果表明，本方法具有较好的准确性。3)采用维纳滤波进行图像复原。根据模糊角度及模糊尺度参数确定点扩散函数，利用点扩散函数在频域中对模糊图像进行逆滤波，得到清晰化图像。实验结果表明，本文方法具有较好的去模糊效果。4)采用Haar小波进行图像去噪。选择Haar小波进行图像分解，滤除高频噪声，再重构得到去噪后的图像。实验结果表明，对去噪图像进行图像复原，能有效抑制复原图像的振铃效应。论文就图像去模糊的关键问题，模糊参数的估计展开，将新的方法用于图像复原，小波去噪抑制振铃效应，提高图像清晰度，改善图像质量，增强鲁棒性具有非常重要的意义。更多还原

【Abstract】 With the development of science and technology, the requirement for imagequality is higher than before. In particular condition, we cann’t get very clear imagefrom the image acquisition device. There are many factors influencing the imagequality, such as the low quality of image acquisition device. They result in themissing of image detail. Besides, the light intensity may lead to the reduction ofimage contrast. Our purpose is to get high quality image that contains large numberof detais from the degraded image. In practical application, it’s more important toextra detail and edge information. We want to get the license information thatoffends vehicle, or the disease information. Thus, it’s necessary to improve imagequality by restoration of the degraded image.Digital image restoration aims at constructing the original image from thedegraded image. Motion blur is very common in various degraded pattern. Whenthere is relative motion between the image acquisition device and the object, themotion blur occurs. It results in the degrade of image quality. We need to clearimage by programming to get clear image. The thesis analyzes the causes of motionblur, and builds the degraded model. Then the estimation of motion blur parametersis introduced in the model, that contains blur angle and length estimation. At last,Wiener Filter is adopted for restoration. With the appearance of Ringing Effect, Thewhole image deblur process is completed. Analyze the causes of Ringing Effectin the image restoration, we conclude that noise is the main factor. In the process ofobtaining、transporting and saving image, noise is inevitable. Thus, haar wavelet isused for denoising in image pre-processing stage to suppress the ringing effect.The image denoising and deblurring are completed in the thesis, and theconcrete work are as follows:1) Gabor filter is adopted to estimate the blur angle. First,16Gabor filter isconvoluted with the image spectrogram. Then according to the angle of thelargest response, determine the direction of every point. The low-energy part isset to zero. At last, the average angle of non-zero points is taken as the motionblur angle;2) Autocorrelation function is used to calculate the blur length. First, Thedifferential image is computed in the level and then the vertical direction. Thenin the horizontal direction, the autocorrelation function of the differential imageis obtained. Add all the autocorrelation curves, and get the sum autocorrelation curve. According to the distance between the symmetric negative peak of thecurve, the blur length is calculated. Experimental results show that the methodis accurate;3) The Wiener filter is used for image restoration. First, the point spread functionis determined by the blur angle and length. In the frequency domain, Inversefilter is used to restore clear image. Experimental results show that therestoration is very good.4) Haar wavelet is adopted to denoising. Based on the wavelet decomposition andreconstruction principle, we choose haar wavelet to decomposite image. Thenthe high-frequency part is filting. At last reconstruct the image which the noiseis get rid of. Experiments show that in image pre-processing stage, the metodcan suppress the ringing effect created by Wiener filter restoration veryeffectively.The thesis began with blur parameter estimation, which is the key issue. A newmethod for image restoration is proposed that is based on Gabor filter. Waveletdenoising is used to damp the ringing effect. By these processing, the image clarityand quality are improved greatly, and the robustness is enhanced.更多还原