一类未知噪声模型的图像去噪方法  

Image Denoising Method with Unknown Noise Model

作　　者：彭宏[1] 赵海英[2,3] 黄甜甜[3] 

机构地区：[1]新疆师范大学网络教育学院,乌鲁木齐830054 [2]新疆师范大学计算机科学技术学院,乌鲁木齐830054 [3]北京科技大学信息工程学院,北京100083

出　　处：《计算机系统应用》2011年第12期205-210,共6页Computer Systems & Applications

基　　金：新疆维吾尔自治区自然科学基金(2011211A029)

摘　　要：图像中包含噪声不仅会降低图像质量,而且严重影响后续相关算法的有效性。高效稳健的去噪方法对于各类信号处理非常重要。为了改善实际夜间远程拍摄图像的质量,引入五种图像降噪方法。首先在合理推导噪声模型的基础上,引入Kalman滤波器去除夜空图像背景噪声;然后分别进行中值滤波、均值滤波、维纳滤波和无参估计的均值漂移算法去除实际夜空图像噪声;最后分析比较五种去噪方法,并给出不同算法的信噪比与峰值信噪比。实验结果:五种降噪方法虽不同程度地降低了夜空图像噪声影响,但均值漂移算法较好地保持了图像有用信息和边缘特征,而且算法不需考虑噪声模型。结论:均值漂移算法展示出在这一类未知噪声模型的夜空背景降噪中的优势。Noise of image does not only reduce the quality of image but also interferes with the validity of correlative processing arithmetic seriously. Therefore, effective and robust methods of removing noise are very important for various signal processing. To improve quality of the actual distance remote control image in the paper, MeanShift algorithm of no parameter estimation is introduced and five methods of removing image noise are compared. Firstly, based on the reasonable assume to be noise model to remove image noise. Kalman filtering is used to remove noise. Then median filtering, mean filtering and Wiener filtering are performed separately. Finally, MeanShift algorithm is applied to remove noise. Experimental results show the five methods which are used in this paper reduce the noise in the night sky image to varying degrees. Moreover MeanShift algorithm in image denoising keeps the detail information and edge character of the image better. Compared with the five traditional filtering methods, MeanShift algorithm shows the advantage in image denoising of the actual night sky image background noise.

关 键 词：未知噪声模型 均值漂移 噪声均方估计 KALMAN滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]