【作者】 毕雪；

【导师】 陈向东；

【作者基本信息】 西南交通大学 ， 通信与信息系统， 2007， 硕士

【摘要】 由于器件和环境的影响，使各类图像信号存在着噪声，对提取图像边缘造成困难。尽管目前有不少关于边缘提取的算法，但在噪声存在的情况下提取边缘并不理想。在时频域里，小波变换具有独特的多尺度分析技术，使信号与噪声在多尺度分解下表现出来的性质不同，信号相邻分解尺度间具有相关性。本文提出一种改进的边缘检测算法—基于小波层间系数反向移位相关提取含噪图像边缘：把图像进行小波多尺度分解，让相邻小波系数反向移位相乘从而增强边缘，形成滤波图象后对某层子图像滤波从而达到去除噪声的效果。根据仿真结果表明该算法比传统边缘提取方法对噪声有更好的抑制作用。随着现代信息技术的发展，人们通过传感器获取含有信息的数据，然而传感器检测到的往往是多个未知信源混在一起的信号，对混合的盲信号进行分离是必不可少的。现有许多独立分量分析(ICA)算法引入了迭代，计算量很大，可能出现不收敛的结果。许多ICA算法仅对超高斯源信号或者亚高斯源信号二者之一有效，但对两者并存的情况缺乏实用性。本文将信号经过小波分解以后的近似系数和细节系数参与到ICA的寻优过程中，以盲源分离效果越好信噪比越高的特点，建立信噪比目标函数。小波近似系数表征了信号的绝大部分信息，提出基于小波变换的全局最优盲分离算法：在建立目标函数以后，把近似系数作为对源信号的估计，寻优过程转换为特征值求解。通过仿真结果表明该算法无需迭代，复杂性和运算较低，分离效果较好。小波细节系数也是信号重要组成部分，在此基础上提出了一种改进的算法：基于小波平滑的超高斯与亚高斯信号盲分离算法。通过仿真结果表明该改进算法具有前一种算法的优点，而且比前一种算法具有更好的相似系数和分离效果。更多还原

【Abstract】 In the influence of apparatus and environment, there are some noise in every kind of images,so it is difficult to detect edges. There have been some algorithms on edge detection,but they can’t get ideal results as noise in them.Wavelet transform provides special multiscales decomposition technology in time frequency domain,making signals and noise show different property and making adjacent scales’ correlation.This paper present an improved edge detection algorithms-edge detection in noisy image based on wavelet interscales back shifting correlation : it resolves image by wavelet backing shiftingly multiply between adjacent wavelet coefficients to enhance edges , then form filtered image and filter some subimages to denoise. Simulation result shows this improved algorithm can better restrain noise than traditional edge detection algorithms.As the development of information technology ,human beings receive data containing information by sensors ,which are mixed by unknown sources. So it is necessary to separate mixed blind sources. Many independent component analysis(ICA) algorithms need iteration and large couting,which may be get divergence result. They are valid when sources are supergaussian or subgaussian,but can’t be practical when mixed souces are supergaussian signal andsubgaussian signal toghter.After wavelet decomposition of signal, it gets approximated coefficients and detailed coefficients which are joined in the ICA optimization process.Then it builts a signal to noise ratio objective function based on peculiarity that better blind source separation result, higher signal to noise ratio.Most imformation of signal is shown by wavelet approximated coefficients,so this paper present ICA method with global optimal property based on wavelet transform: after building objective function ,let approximated coefficients as estimation of source signals, optimization is changed into eigenvalue decomposition.This method haven’t any iteration,low complexity and counting from simulation result. Wavelet detailed coefficients are important part of signal,so it presents an improved algorithm: blind separation in supergaussian and subgaussian signals based on wavelet smoothing. Simulation result shows this improved method have same merit of preceding method, but better resemblant coefficients and separated effect.更多还原