运动图像去模糊算法研究

发布时间：2018-04-01 23:01

本文选题：运动模糊　切入点：参数估计　出处：《广西师范大学》2017年硕士论文

【摘要】：图像在拍摄过程中会因为相机的抖动或者是物体本身的运动产生运动模糊。运动图像去模糊的目的是从模糊图像中恢复出清晰的图像信息。在模糊核未知和模糊图像含噪声的情况下,图像去模糊难度大大加深。本文在介绍了图像模糊的退化模型等理论基础后,对模拟图像和实拍图像进行去模糊算法研究。本文的工作主要在以下几个方面:(1)研究了改进的自适应中值滤波方法对椒盐噪声进行去噪,此方法仅对疑似噪声进行中值滤波,弥补了常规中值滤波造成图像细节丢失的缺陷。(2)研究了一种改进的匀速直线运动估计模糊核的方法。模糊角度的估计采用改进的基于二次傅立叶变换和霍夫变换的模糊角度估计法。设计了一个改进的自适应模糊长度的估计方法,首先对模糊长度进行自相关估计,当所求模糊长度小于所设阈值时再用频谱投影法进行估计。设计精确估计的方法对所估计的模糊核进行误差范围内的遍历,把所遍历的值进行维纳滤波去模糊算法检验,精确估计出最佳的模糊核。计算机仿真实验结果表明:此方法能很好的估计匀速直线运动模糊核。此研究为后续运动图像去模糊算法研究打下了较好的前期基础。(3)研究了一种基于遗传自适应的维纳滤波去模糊方法。此方法用遗传算法自适应优化维纳滤波的正则化项k值达到最佳图像复原效果。本方法不仅适用于含噪声和不含噪声的模拟图像去模糊,对实拍图像去模糊也具有一定的效果。(4)研究了一种改进的自适应中值滤波的RL图像去模糊方法。对于含噪图像,先对噪声进行改进的自适应中值滤波然后再用RL算法对图像去模糊。之后对实拍图片进行两种改进的方法去模糊效果的比较,依据计算机仿真结果对两种方法的最佳使用范围进行了讨论。(5)设计了以MatlabGUI为平台的运动图像去模糊系统。
[Abstract]:The image will be blurred by the camera jitter or the motion of the object itself.The purpose of motion image deblurring is to recover clear image information from blurred image.When the fuzzy kernel is unknown and the blurred image contains noise, the image deblurring becomes more difficult.In this paper, the deblurring algorithm of simulated image and real image is studied after introducing the theory of image blur degradation model and so on.The main work of this paper is as follows: 1) the improved adaptive median filtering method is studied to remove salt and pepper noise. This method only applies median filtering to suspected noise.This paper makes up for the defect of image detail loss caused by conventional median filter. (2) an improved method for estimating blur kernels of uniform linear motion is studied.An improved fuzzy angle estimation method based on quadratic Fourier transform and Hough transform is used to estimate the fuzzy angle.An improved self-adaptive fuzzy length estimation method is designed. Firstly, the fuzzy length is estimated by autocorrelation. When the fuzzy length is less than the threshold, the spectral projection method is used to estimate the fuzzy length.The method of accurate estimation is designed to ergodicate the estimated fuzzy kernel within the error range, and the ergodic value is verified by Wiener filter de-fuzzy algorithm, and the best fuzzy kernel is estimated accurately.The computer simulation results show that this method can estimate the uniform linear motion blur kernel well.This study lays a good foundation for the further study of deblurring algorithm for moving images. (3) A genetic adaptive Wiener filter de-blurring method is studied.The genetic algorithm is used to adaptively optimize the regularization term k of Wiener filter to achieve the best image restoration effect.This method is not only suitable for deblurring simulated images with and without noise, but also has a certain effect on de-blurring of real images.) an improved adaptive median filtering method for de-blurring of RL images is studied in this paper.For noisy images, an improved adaptive median filter is applied to the noise and then the RL algorithm is used to deblur the images.Then two improved methods to deblur images are compared. According to the results of computer simulation, the optimal application range of the two methods is discussed. (5) A motion image deblurring system based on MatlabGUI is designed.
【学位授予单位】：广西师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】