遥感图像条带噪声的多尺度变分模型去除

发布时间：2018-03-30 04:16

本文选题：光学遥感　切入点：条带噪声　出处：《光学精密工程》2017年01期

【摘要】：多片CCD拼接遥感成像系统由于存在非均匀性问题,导致遥感图像中常存在条带噪声,本文在分析条带噪声的主要来源和模型的基础上,提出了多尺度变分模型的条带噪声去除方法。首先,分析了条带噪声的特点并建立了图像退化模型。其次,结合条带噪声的单向性特点与多尺度分层分解方法构造能量泛函。然后,利用不动点Gauss-Seidel迭代法多尺度分级极小化能量泛函,将条带噪声和图像有用信息分离。最后,对各尺度结构分量和细节分量进行累加,得到去噪图像。实验结果表明:对于周期条带噪声,图像畸变量为2‰,图像辐射质量提升到11.715dB;对于随机条带噪声,图像畸变量为3.3‰,图像辐射质量提升到11.092 5dB。与典型条带噪声去除方法相比,不管是周期条带噪声还是随机条带噪声,本文方法均能够在保证畸变量很小的情况下,将其完全去除,满足遥感图像低畸变量的预处理要求。
[Abstract]:Multi CCD mosaic remote sensing imaging system due to the presence of non uniformity of the resulting band noise often exist in remote sensing image, based on the analysis of the main sources of stripe noise and model, proposed a multiscale variational model destriping method. Firstly, analyzes the characteristics and the establishment of band noise the image degradation model. Secondly, combined with the band noise characteristics of unidirectional and multiscale hierarchical decomposition method to construct the energy function. Then, by using the fixed point iterative method of Gauss-Seidel hierarchical multiscale minimization energy functional, the band of noise and image information separation. Finally, the accumulation of scale structure and detail components are. Image denoising. Experimental results show that the periodic stripe noise, image distortion is 2 per thousand, to enhance the quality of radiation image 11.715dB; for the random band noise, image distortion of image is 3.3. To improve the quality of radiation to 11.092 5dB. with typical destriping compared to either periodic band noise or random band noise, this method can ensure the distortion is small, remove it completely, to meet the low distortion of remote sensing image preprocessing requirements.

【作者单位】：中国科学院长春光学精密机械与物理研究所;中国科学院大学;
【基金】：吉林省科技发展计划资助项目(No.20126016)
【分类号】：TP751

【相似文献】