【作者】 张智；

【导师】 夏德深；

【作者基本信息】 南京理工大学 ， 模式识别与智能系统， 2008， 博士

【摘要】 倒易晶胞的复原方法是一种面向遥感图像的单帧图像复原方法。该方法通过在频域网格内去除或减少相邻频谱网格的频谱交叠,以达到去除混迭的目的。再结合传统的图像复原方法,进行去噪声和去模糊,最后恢复一幅较为清晰的图像。本文对倒易晶胞的复原算法进行深入的研究,主要有以下几方面的工作:研究了遥感图像中混迭现象产生的原因,进行混迭量对图像影响的定量化测试。通过混迭进行量化,测试混迭量对Wiener结合倒易晶胞复原和TV(Total Variation)结合倒易晶胞方法的影响。并针对实验中的具体图像,限定倒易晶胞复原方法能够处理的混迭程度。从对MTF(Modulation Transfer Function)补偿的角度分析,倒易晶胞的物理意义就是一个低通滤波器。它在滤波中的作用就是既尽量趋向理想的Shannon滤波器,又要使得倒易晶胞形状保持光滑。研究倒易晶胞中去卷积的方法,通过实验发现倒易晶胞复原工作中可能存在的问题,即彩色混迭和彩色噪声现象。首次提出彩色混迭的概念,并给出彩色混迭的公式。从实验结果中,给出了彩色混迭和彩色噪声在倒易晶胞中的表现。利用频域结合时频域,提出一种基于小波理论结合倒易晶胞的复原方法,把混迭考虑到数学模型中。既达到了去混迭的目的,又可得到比直接频域复原方法更好的复原效果。研究了亚像元技术及其插值方法,提出一种基于两帧亚像元图像并结合中值滤波的小波复原方法,达到较好的复原效果。在此基础上进一步研究,又提出一种基于复小波的亚像元复原方法。该方法利用复小波进行插值和双变量层间去噪,与基于两帧亚像元图像的小波复原方法相比得到更好的复原效果。更多还原

【Abstract】 Reciprocal cell restoration was based on single frame of satellite image. The restoration was transformed image into frequency domain. The superposition part of the frequency spectrum was removed between adjacent grids by proposed method. The conventional restorations were combined with the reciprocal cell to denoise and deblur. Finally, the clear image was obtained. This dissertation carried out a further study on the reciprocal cell theory and its restorations of remote image. Study has been done in the following aspects:Firstly, the cause of the aliasing on remote sensing image was analyzed. In addition, the distortion of aliasing was tested quantificationally on images. The distortion of aliasing was special for the Wiener and TV(Total Variation) combined with reciprocal cell. The extent for the range of aliasing was defined through specific example images.From the viewpoint of MTF(Modulation Transfer Function) Compensation, the physical meaning of reciprocal cell was a low-passed filter. It trends to Shannon filter and retained figure smooth at the same time.Besides, the deconvolution of reciprocal cell was studied. There exsited phenomena in the proceeding of deconvolution, such as colour noise and colour aliasing. And the concept of colour aliasing was raised creatively here in the first time. The formula of colour aliasing was derived unprecedentedly. Whereafter, the manifestation of colour noise and colour aliasing were brought forth on frequency domain.Moreover, the restoration of wavelet combined with reciprocal cell was proposed. The mathematic model was formulated with aliasing. The hybrid of Wavelet- Reciprocal cell could dealias, denoise and deblur, which was better than restorations on frequency domain.Finally, the subpixel technique and its interpolations were studied. The wavelet interpolation of subpixel images and median filter was brought forward. After further study, the complex wavelet of subpixel technic was raised. The method was the hybrid between complex wavelet interpolation and denoising of dual-trees complex wavelet. Thereafter, test showed this algorithm could receive better results than the wavelet interpolation.更多还原