成像系统非均匀性条带变分正则化校正方法研究-模式识别与智能系统专业论文.docx

进一步，针对图像在获取的过程中经常遭受条带噪声及其它各种各样的混合噪 声：高斯随机噪声，死像素等问题，为了节省处理时间，本文提出一种统一的图像 降噪方法来同时的去除混合噪声。通过空域内字典学习的手段，本质上利用了图像 内在的自相似性，来剥离随机噪声，同时利用梯度域单方向总变分来抑制条带噪声， 把握住了条带噪声的方向性，达到去除噪声的同时，完美的保存好图像固有的结构 信息。 最后，鉴于传统的遥感图像条带校正算法大都在单幅图像内处理，忽略并破坏 了相邻谱带间固有连续性，本文提出一种基于谱-空联合的总变分条带校正算法，将 遥感图像多谱数据看作一个谱-空数据立方体，通过施加一个各项异性的谱-空三维总 变分约束，保证遥感图像在谱-空内在的平滑性，从而达到去除噪声的同时保存好谱- 空结构信息的完整性。 关键词：多传感元成像系统，条带噪声，变分正则化，先验知识，稀疏表达，分裂 Bregman 迭代 Abstract Due to the limitation of the detectors, multidetector imaging systems often suffer from the problem of stripe noise, random noise and blurs, which greatly degrade the imaging quality, leading to difficulty in succeeding image interpretation processes. The main cause of stripe noise is the difference in the response of detectors. The difficult of stripes correction lie on how to effectively remove different kinds of stripes, and well preserve the detail information. This paper focuses on stripe correction problem, including the analysis of the characteristic of the stripe noise, the unified modeling of the degrade process, the sparsity representation of the images in spatial-spectral-transform domain and also fast optimization methods. By making use of the variational regularization technique, our dissertation mainly includes the following aspects: First, we analyse the characteristic of the stripe noise, and construct the unified degraded model which can reflect the influence of the stripe, random noise and blur. Most of the existing algorithms were designed for specific stripe noise in certain detector. However, the stripes caused by different detectors contain obviously different characteristics. Therefore, the existing algorithms lack of good roubustness when they face different multidetector imaging systems, which makes them impractical in real applications. This dissertation firstly analyses the degradation of the stripes, and then concludes the structure characteristic of the stripe noise. Therefore, we construct the unified the mathematic degradation model of