高分辨率太阳图像中列固定模式噪声的消除

发布时间：2018-06-21 08:28

本文选题：CMOS传感器 + 列固定模式噪声　；参考：《昆明理工大学》2017年硕士论文

【摘要】：以小体积,功耗低,抗辐射等优良特性著称的CMOS传感器为高分辨率太阳图像的采集提供了可靠的支持。而高分辨率图像的获得为太阳大气,太阳物理等科学研究提供了有力保障。但是,由于CMOS传感器处理电路和模数转换器之间的失配,传感器之间的差异等会导致图像数据中出现列固定模式噪声(CFPN)。这些噪声的存在不但降低了图像的质量,掩盖了图像细节,而且对后期围绕这些图像展开的科学研究也产生了影响。因此,列固定模式噪声的去除是非常有必要的。近年来,国内外众多学者根据列固定模式噪声的分布特性和统计规律,提出了基于统计学,变分,傅里叶结合滤波以及多尺度去噪法等多种算法。但是由于这些算法自身的局限性使去噪后的图像出现模糊失真等现象。为了解决上述问题,本文在小波变换的基础上提出了一种基于小波变换和双滤波的去噪算法。本文首先以仿真实验的方式展示算法执行过程。算法的主要执行过程为:首先,根据噪声的产生机理以及存在形式,将原始图像对数化并进行小波变换。其次,对小波域中的垂直分量进行中值滤波,去除其中的噪声小波系数。接着,利用小波逆变换得到无噪图像,并与对数化图像做差提取初始噪声。然后,对初始噪声进行低通高斯滤波并指数化得到结果噪声。最后,用原始图像除以结果噪声便得到消噪后的图像。同时,本文选取了几个目前已存在的固定模式噪声消除方法与本文算法作对比,验证所提算法的准确性和有效性。仿真结果表明,所提算法能够消除94%的噪声。对均值、峰值信噪比(PSNR)、结构相似度(SSIM)和功率谱密度等评价指标的分析表明,本文算法可以得到更佳的消噪结果。为了检验算法对阈值参数的响应程度,针对仿真实验分别对比和分析了不同阈值情况下的消噪结果差异。分析结果表明所提算法对中值滤波窗口宽度以及高斯核的选取有较强的反映。利用本文算法对中国云南抚仙湖天文台TiO波段以及美国大熊湖天文台Ha波段2组数据进行消噪处理,所得结果表明,本文算法不但能够准确去除噪声,而且能够最大限度的保留图像细节信息,消噪后的图像特征更加明显。
[Abstract]:The CMOS sensor, known for its small size, low power consumption and radiation resistance, provides reliable support for the acquisition of high resolution solar images. The acquisition of high resolution images provides a powerful guarantee for the scientific research of the solar atmosphere and solar physics. However, because of the mismatch between the CMOS sensor processing circuit and the analog to digital converter, The difference between sensors will lead to the appearance of column fixed pattern noise (CFPN) in the image data. The existence of these noises not only reduces the quality of the image, covers the details of the image, but also has an impact on the later scientific research on the expansion of these images. Therefore, the removal of the column fixed pattern noise is very necessary. In order to solve the above problems, many scholars at home and abroad have proposed a variety of algorithms based on statistics, variation, Fourier combined filtering and multiscale denoising based on the distribution and statistical rules of the fixed pattern noise. But because of the limitations of these algorithms, the images after the denoising are blurred. On the basis of wavelet transform, a denoising algorithm based on wavelet transform and double filter is proposed. Firstly, the execution process of the algorithm is displayed in the simulation experiment. The main execution process of the algorithm is: firstly, the original image is logarithmic and wavelet transform is carried out according to the mechanism of noise generation and the existence form. Secondly, the wavelet domain is applied to the wavelet domain. The vertical component is filtered by the median filter to remove the noise wavelet coefficients. Then, the noise free image is obtained by using the wavelet inverse transform, and the initial noise is extracted from the logarithmic image. Then, the initial noise is reduced by the low pass Gauss filter and exponentially obtains the result noise. Finally, the noise is eliminated with the original image and the noise is de-noised. At the same time, a few existing fixed mode noise elimination methods are selected and compared with the algorithm in this paper to verify the accuracy and effectiveness of the proposed algorithm. The simulation results show that the proposed algorithm can eliminate 94% noise. Value, peak signal to noise ratio (PSNR), structural similarity (SSIM) and power spectral density are evaluated. The analysis shows that the algorithm can get better noise elimination results. In order to test the response of the algorithm to the threshold parameters, the difference of noise elimination results under different threshold conditions is compared and analyzed. The results show that the proposed algorithm has a strong reflection on the width of the median filter window and the selection of the Gauss kernel. The algorithm is used to denoise the data of 2 groups in the TiO band of the Yunnan Fuxian Lake Observatory in China and the Ha band of the Big Bear Lake Observatory in the United States. The results show that the algorithm can not only remove the noise accurately, but also retain the details of the image to the maximum, and the image features after the noise elimination are more obvious.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212

【参考文献】