【作者】 陈波；

【导师】 耿则勋；

【作者基本信息】 解放军信息工程大学 ， 摄影测量与遥感， 2008， 博士

【摘要】 通过大气湍流成像是所有工作在大气环境中的光学成像系统必然会遇到的问题。在光线进入成像传感器入瞳之前,大气介质影响或干扰光波的理想传播,使目标反射电磁波的波阵面产生畸变,形成焦平面上像点强度分布扩教、峰值降低以及光束漂移等湍流效应,最终导致图像模糊,严重地影响光学系统的成像性能。自适应光学技术是目前克服大气湍流最有效的方法之一,但其对大气湍流的补偿是不完全的,目标的高频信息仍然受到抑制和衰减。因此,对经过自适应光学校正后的图像还必须进行基于图像复原技术的后处理,才能获取更清晰的目标图像。本文主要针对我国自适应光学地基望远镜成像后处理的迫切需求,利用信号处理和计算机软件技术对图像进行高清晰复原,以消除自适应光学校正的残余误差,使其接近或达到成像系统的衍射极限。论文重点研究了自适应光学图像复原理论与算法,同时对自适应光学波前重构和图像复原质量评价等问题也进行了探讨与分析,完成的工作主要包括:1.开发了自适应光学成像原理演示系统。根据光波大气湍流传输理论,分别分析了湍流对长曝光OTF和短曝光OTF的影响。在介绍自适应光学系统组成和工作原理的基础上,基于SCIAO平台设计一个简单的自适应光学成像原理演示系统CYAOIS。2.定义了图像复原质量综合评价指数。针对现有图像复原质量评价方法没有考虑伪像的不足,根据人眼视觉的特点,定义了一种基于分区局部方差变化统计的图像复原质量综合评价指数。实验证明,该指数是无参考图像复原质量评价的一种客观有效的指标。3.基于广义岭估计的自适应光学图像波前重构。从自适应光学图像后处理复原的角度来讲,实时性和快速计算不再是波前重构的基本要求,而高精度才是波前重构最重要的要求。按照时间换精度的思路把广义岭估计GRE和迭代答解的思想引入波前重构,提出一种面向图像复原的广义岭估计Zernike模式波前重构算法。实验证明,该算法重构的PSF在图像复原中有更好的图像重建效果。4.非抽样小波变换降质图像PSF估计。在图像复原问题中,PSF估计是算法的核心部分。把非抽样小波变换引入降质图像PSF估计,在Fried参数模型的基础上,根据不同尺度下小波变换模极大值和高斯点扩展函数方差的关系,提出一种基于非抽样小波变换的PSF估计算法。实验结果证明,该算法对无波前测量数据自适应光学图像有较好的PSF估计精度。5.基于ENAS-RIF算法的自适应光学图像复原。针对基本NAS-RIF算法的不足,把Curvelet去噪图像预处理、可靠支持域、目标结构约束项和增强代价函数引入NAS-RIF算法,提出了基于可靠支持域和改进代价函数的ENAS-RIF算法。自适应光学图像复原实验结果证明,ENAS-RIF算法较之NAS-RIF算法有更快的收敛速度、更好的恢复效果。6.基于IRL-IBD算法的自适应光学图像复原。在非对称IBD算法的基础上,本文在PSF频率域引入带宽有限约束来进一步提高算法的可靠性,在PSF像素空间引入动态支持域的思想以加快算法收敛速度,提出多重约束非对称IRL-IBD算法用于自适应光学图像高清晰复原。实验结果证明,改进后的IRL-IBD算法复原性能明显优于IBD算法。7.基于方差统计的图像序列不良帧剔除。在多帧图像高清晰复原中,一些降质严重的观测图像不但不能对复原作出贡献,而且将可能严重影响重建图像的质量。本文利用加性噪声导致降质图像方差增大和PSF模糊导致降质图像方差减小的性质,参考摄影测量中粗差剔除的思想,设计了一个基于图像方差统计的不良帧自动剔除方案。实验结果表明,剔除不良帧后的图像复原效果有明显改善。8.基于MAP原理的自适应光学图像多帧MF-MAPJD高清晰复原。在Bayesian框架下,根据自适应光学图像的特点重新定义了混合噪声模型和目标结构先验模型,在代价函数中新增加了PSF准则项,设计了一个简单的正则化约束项自动平衡方案,提出了基于MAP原理的AO图像多帧联合解卷积算法MF-MAPJD。实验结果表明,新算法同时估计点扩教函数和目标,充分了利用图像序列中的各种先验约束信息,复原图像质量改善明显。9.基于二代Curvelet变换的图像自适应去噪。利用Curvelet变换比小波变换对图像信号具有更稀疏表示的特点,结合Bayesian Shrink理论,提出了一种改进的Curvelet域图像自适应去噪算法。实验结果表明其去噪性能优予小波去噪算法。10.基于二代Curvelet变换的自适应光学图像复原。把多尺度多方向的思想引入自适应光学图像处理,设计了基于二代Curvelet变换的Fourier-Curvelet混合域正则化图像复原算法ForCuRD。ForCuRD算法同时采用了傅里叶域收缩和Curvelet域收缩,克服单一变换域收缩的局限,为自适应光学图像复原研究提供了一条新思路。更多还原

【Abstract】 All the optical imaging systems which work in atmospheric circumstance will face the problem of imaging through atmospheric turbulence.Before the light comes into the focus of imaging sensor, atmospheric medium may affect or disturb the ideal transmission of light,which will cause several effects, such as the aberrance of wave front of object’s reflecting electromagnetic wave,the diffusion of image point intensity distribution in the focal plane,the reduce of peak value and the excursion of light beam.In the end,it will cause the blurring of image and badly impact the imaging capability of optical systems. The technology of adaptive optics is one of the most effective methods of overcoming atmospheric turbulence.But it cannot completely compensate the atmospheric turbulence,and the high frequency information of objects may be bated and reduced.So the image which has been rectified by adaptive optics should be post-processed by the technology of image restoration,and then it will be a clearer object image.This paper mainly aims at the imminent need of imaging post-process of adaptive optics groundwork telescope in our country,and uses the technology of signal process and computer software to restore high-legible image and eliminate the remain error of adaptive optics correction,so it will meet or reach the diffraction limit of imaging system.The paper mainly researches on the theory and algorithm of adaptive optics image restoration and at the same time discusses and analyzes the problems of adaptive optics wave-front reconstruction and the evaluation of image restoration quality.The main work include as follows:1.It developed a demo system of adaptive optics imaging theory.According to the theory of wave’s atmospheric turbulence transmission,it analyzes the effect of turbulence on long-expose OTF and short-expose OTF.Based on the presentation of adaptive optics system’s constitution and work theory,it designed a simple demo system CYAOIS of adaptive optics imaging theory based on SCIAO platform.2.It defined the synthetical evaluation index of image restoration quality.Aimed at the shortage of existing image restoration quality evaluation methods which cannot consider the fake image,it defined a synthetical evaluation index of image restoration quality based on subarea local variance change statistic according to the human eyes vision characteristic.The experiment proved that this index is an effective and objective index to evaluate non-reference image restoration quality.3.The wave-front reconstruction of adaptive optics image based on generalized ridge estimation.In the view of adaptive optics image post-processed restoration,real time and fast computation are not the basic request of wave-front reconstruction,but the high-precision.On the consideration of sacrifice time for precision,it brought the idea of generalized ridge estimation(GRE) and iterative solution into wave-front reconstruction,and presented a wave-front reconstruction algorithm of generalized ridge estimation’s Zernike mode based on image restoration.The experiment proved that the algorithm which restored PSF had performed well in image restoration.4.PSF estimation of degraded image based on non-sample wavelet transformation.In the problems of image restoration,PSF estimation is the main part of the algorithm.It brought non-sample wavelet transformation into the PSF estimation of blurred image.Based on the Fried parameter model and the relationship between module max value of wavelet transformation and the variance of Gauss point spread function in different scale,it presented a PSF estimation algorithm based on the non-sample wavelet transformation.The experiment proved that the algorithm had well PSF estimation precision in adaptive optics image without pre-wave survey data.5.Adaptive optics image restoration based on ENAS-RIF algorithm.Aimed at the shortage of ENAS-RIF algorithm it brought image pre-process of Curvelet denoise,credible support region, object construction restriction item and enhanced cost function into NAS-RIF algorithm And it presented ENAS-RIF algorithm based on credible support region and enhanced cost function.The experiment of adaptive optics image restoration proved that the convergence speed of ENAS-RIF algorithm was faster than NAS-RIF algorithm and it did well in image restoration.6.Adaptive optics image restoration based on IRL-IBD algorithm.On the basis of non-symmetry IBD algorithm,it improved the reliability of algorithm by introducing the bandwidth finite restriction into PSF frequency domain and the convergence speed of algorithm by introducing dynamic support region into PSF estimation,and it presented multi-restriction non-symmetry IRL-IBD algorithm into adaptive optics image restoration.The experiment proved that reformative IRL-IBD algorithm was better than IBD algorithm7.Eliminating ill frame of image sequence based on variance statistic.In the restoration of multi-frame image,some observation images with bad quality cannot contribute to the image restoration,but may badly affect the quality of restored images.It used the characteristics that additive noises may increase the variance of blurred image and PSF blurring may decrease the variance of blurred image, and considering the idea of outliers detection in photogrammetry it designed an automatic detection scheme of ill frame based on image variance statistic.The experiment proved that the image restoration was better after the detection of ill frame.8.The MF-MAPJD high-resolution restoration of multi-frame adaptive optics image based on MAP.In the frame of Bayesian,according to the characteristic of adaptive optics image,it redefined the mixed noises model and the priori model of object construction,and then added PSF restriction item into cost function,and designed a simple self-poise scheme of regularization restriction.And it presented a multi-frame combination deconvolution algorithm MF-MAPJD for AO image.The experiment proved that the new algorithm considered both the point spread function and object,and made the best of all the priori restriction information in the image sequence,and it proved that the quality of image restoration was better.9.Adaptive denoise of image based on the second generation Curvelet transformation.According to the trait that Curvelet transformation is sparser than image signal,and combined with Bayesian Shrink theory,it presented an improved adaptive image denoise algorithm of Curvelet domain.The experiment proved that it was better to denoise than wavelet algorithm.10.Adaptive optics image restoration based on second generation Curvelet transformation It brought multi-scale and multi-direction idea into image process,and considered the idea of image Fourier-Wavelet hybrid domain regularization deconvolution,and then it proposed the image Fourier-Curvelet hybrid domain regularization deconvolution(ForCuRD) based on second generation Curvelet transformation.ForCuRD algorithm used both the Fourier domain constriction and Curvelet domain constriction,and overcame the shortage of single transformation domain constriction,and lend a new way for the research of adaptive optics image restoration.更多还原