无人机遥感模糊图像恢复技术研究

发布时间：2018-01-10 20:19

本文关键词：无人机遥感模糊图像恢复技术研究　出处：《中国科学院长春光学精密机械与物理研究所》2017年博士论文　论文类型：学位论文

【摘要】：无人机遥感成像可以经济、快速、安全的获取地面信息,所以在资源勘探、环境监测、战场侦察等领域具有很高的应用价值。由于无人机航拍成像时受到恶劣天气、发动机振动、自身倾斜晃动、相机相对运动和大气扰动的影响,所捕捉到的图像具有噪声复杂、模糊、对比度低、细节纹理不清等特点,使图像质量大大下降。为获取清晰图像信息,现今大多数图像恢复算法都是建立在模糊图像点扩散函数已知的前提下,如逆滤波、维纳滤波等,但现实中的成像条件却很复杂。航空相机自身的不规则震动、与所拍目标间的相对运动、大气湍流等都是未知的,无法精确得知模糊核函数,因此对盲复原算法的研究,有着现实和理论的双重需求。论文针对运动模糊、相机噪声和异常值干扰模糊核估计、大气传输模糊产生的退化图像复原问题,分别建立优化函数模型并设计相应的复原方法,以去除图像退化现象、提高图像质量与保留图像细节信息等因素为主要切入点研究无人机遥感模糊图像复原的新方法,其主要内容如下:针对无人机航拍时的运动模糊问题,提出了一种基于L0稀疏先验的无人机遥感图像复原算法。首先,通过对遥感图像特性进行分析,得出了运动模糊图像的梯度分布要比清晰图像稠密并且暗通道的稀疏性也相对较小这一固有属性,建立了新的L0稀疏正则化复原模型。接着,针对L0范数的高度非凸性和暗通道稀疏优化过程中涉及到的非线性最小化问题,利用查表法构建了一种近似线性映射矩阵,并用半二次分解法对L0最小化问题进行求解。最后,采用快速傅里叶变换在频域中对模糊核及清晰图像进行交替迭代运算,得出复原图像。实验结果表明,运动模糊图像得到了有效恢复,各项图像质量客观评价指标均有显著提升,并可以有效抑制图像边缘处的振铃效应,完整保留清晰细节信息的同时显著提高了运算速度。由于航空相机异常值及非高斯噪声的存在,严重影响了模糊核的正确估计,使航拍图像恢复效果不佳、细节丢失严重、人工痕迹明显。为此,提出了一种基于消除相机异常值的饱和模糊图像盲复原算法。首先,根据饱和图像灰度特性建立L1正则化模型,引入超拉普拉斯先验提取图像显著边缘。接着,针对S型函数无法完全滤除边缘中的饱和像素,提出一种模糊核镜像辅助函数,通过设定阈值可以有效消除异常值。最后,分析异常值对模糊核估计的影响,建立基于异常值感知的盲反卷积模型,采用迭代加权最小二乘法运算得到恢复图像避免了迭代求解中的二次型问题。实验结果表明,该算法可以极大地降低航空相机异常值的影响,正确估计模糊核函数,优于传统的图像盲复原算法。最后针对无人机遥感航拍图像在获取过程中受到大气扰动影响产生的大气模糊降质问题,提出了一种基于多次散射APSF估计的大气退化图像恢复算法。该方法通过分析大气对光线散射和吸收的物理特性,构建大气传输点扩散函数估计模型,并设计与该模型相匹配的新算法,旨在去除无人机遥感退化图像的大气扰动模糊,完成该类降质图像的复原。经实验仿真,本文方法与其他的传统算法比较,图像恢复质量更加优秀,并对噪声干扰具有一定的鲁棒性。
[Abstract]:UAV remote sensing imaging can be fast, safe and economic, to obtain the information of ground, so in resource exploration, environmental monitoring, and has high application value field of Battlefield Reconnaissance UAV aerial imaging. Due to bad weather, the vibration of the engine, its tilt camera shake, relative motion and atmospheric disturbance, image capture the noise is complex, fuzzy, low contrast and texture details are not clear, so the image quality is greatly reduced. In order to obtain clear image information, most of today's image restoration algorithm is based on the premise of the known fuzzy image point spread function, such as inverse filtering, Wiener filtering, but the reality has the imaging condition very complicated. Its irregular aerial camera shake, and shoot relative motion between targets and atmospheric turbulence are unknown, not to know precisely the fuzzy kernel function, so the blind restoration algorithm The study has the dual needs of reality and theory. The thesis focuses on the motion blur, camera noise and outliers interference blur kernel estimation, atmospheric transmission blur degraded image restoration problems, establish optimization function model and design the corresponding restoration method, to remove image degradation, improve the quality of the image and preserve the detail information of the image such factors as the main starting point of the research on new method of UAV remote sensing fuzzy image restoration, its main contents are as follows: according to the motion of the UAV aerial fuzzy problem, put forward a kind of UAV remote sensing image restoration algorithm based on sparse prior L0. First of all, based on the characteristics of remote sensing image analysis, the gradient of motion blurred image the distribution of dense and dark images than the sparsity of the channel is smaller than the intrinsic property, established L0 sparse novel regularized restoration model. Then, According to the nonlinear minimization problem highly non convex L0 norm sparse optimization and dark channel involved in the process, using the look-up table method to construct an approximate linear mapping matrix, and solves the L0 minimization problem with quadratic decomposition method. Finally, in the frequency domain of fuzzy kernel and clear images are computed by the alternate iteration fast Fourier transform of the restored image. The experimental results show that the motion blurred image has been effectively restored, the objective image quality evaluation index has significantly improved, and can effectively suppress the ringing effect near the edge of the image should be intact, clear details and significantly improves the speed of operation. Because of aerial camera outliers and non Gauss noise the existence of serious impact on the correct estimation of fuzzy kernel, the aerial image restoration effect is poor, serious loss of details, artifacts. Therefore, put forward A saturated fuzzy image blind restoration algorithm to eliminate outliers based on camera. First of all, based on the L1 regularization model of gray characteristics of saturated images, extracting image edge was the introduction of Chau Laplace prior. Then, according to the S function cannot be completely saturated pixel edge filtering, proposes a fuzzy kernel image auxiliary function, by setting the threshold can be to effectively eliminate outliers. Finally, analysis of the abnormal value of fuzzy kernel estimation, establish the abnormal value of blind deconvolution based on perception model, using iterative weighted least squares method to calculate the restored image to avoid two problems in the iterative solution. The experimental results show that this algorithm can greatly reduce the influence of aerial camera outliers. The correct estimation of fuzzy kernel function, blind image restoration algorithm is better than the traditional. Finally, the UAV remote sensing aerial image to the atmosphere in the acquisition process The quality of the fuzzy drop perturbation of the atmosphere, put forward a kind of multiple scattering APSF estimation of atmospheric degradation image restoration algorithm based on this method. Through the analysis of atmospheric light scattering and absorption properties, construction of atmospheric transmission point spread function estimation model, and designs a new algorithm to match the model, in order to remove the UAV remote sensing image degradation atmospheric disturbance fuzzy, the complete degraded image restoration. The experimental simulation, compared with other traditional algorithms this method, image restoration with excellent quality, and has a certain robustness to noise.

【学位授予单位】：中国科学院长春光学精密机械与物理研究所
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP751

【相似文献】