图像超分辨率盲重建方法的若干关键问题研究

Blind super-resolution (SR) reconstruction techniques have been widely applied in many areas such as medical imaging, objective identification and satellite imaging analysis, and public security recovery and detection. Under unknown warping and blurring parameters, it requires obtaining a high resolution image from a sequence of low resolution images with warping, blur, and noise. Because of its complexity, there are problems to be solved and the imaging reconstruction model, the motion blur identification, and the fast and robust reconstruction algorithm have been key problems to be solved and challenging topic in the world. This project is to solve three key problems. Developing the 2D multi-channel reconstruction model between the high resolution image and the low-resolution image to overcome the computation difficulty in the 1D multi-channel model; studying image registration techniques, propose a noise-constrained image registration approach with robustness; studying 2D-ARMA parameter estimation, propose a 2D-ARMA-based method for fast blurr identification to solve the non-convex problem; studying 2D cost function for image reconstruction based on combining L2 norm and L1 norm, developing a super-resolution image blind reconstruction algorithm based on the 2D model, which effectively solves the problems of huge storage space and long calculation time required in the classical reconstruction process. Finally, we analyize the proposed algorithm convergence and error and show that the proposed new algorithm has a good performance in robustness and fast rate, so that the SR research work can make progress in the viewpoint of theory and application.

图像超分辨率盲重建技术广泛应用于医疗成像、目标识别、卫星图像分析、公安侦破和交通安全检测等诸多领域。它要求在运动变形参数和模糊参数未知情形下,由一组低分辨率图像获取一幅质量好的高分辨率图像。该技术综合复杂，尚有许多要解决的问题，而降质重建模型、运动模糊辩识、快速鲁棒的重建方法是当前研究的关键和重点。申请项目旨在研究解决存在的关键问题。建立二维的超分辨率图像重建模型，解决传统一维重建模型的难以计算问题；研究基于图像灰度与特征的配准方法，提出基于噪声约束均方估计的鲁棒配准算法；研究二维回归移动平均参数估计,提出模糊盲辩识的快速算法，解决非凸的疑难问题；研究基于凸组合2-范数和1-范数度量的最优估计，研究最优次梯度与共轭下降技术，提出基于适应凸组合2-范数和1-范数度量的图像超分辨率盲重建的快速鲁棒算法，并在理论、计算、应用效果上予以证实，推动图像超分辨率重建技术在理论及实时应用上重要发展。

图像超分辨率盲重建技术广泛应用于医疗成像、目标识别、卫星图像分析、公安侦破和交通安全检测等诸多领域。它要求在运动变形参数和模糊参数未知情形下,由一组低分辨率图像获取一幅质量好的高分辨率图像。该技术综合复杂，尚有许多要解决的问题，而降质重建模型、运动模糊辩识、快速鲁棒的重建方法是当前研究的关键和重点。申请项目旨在研究解决存在的关键问题。建立二维的超分辨率图像重建模型，解决传统一维重建模型的难以计算问题；研究基于图像灰度与特征的配准方法，提出基于噪声约束均方估计的鲁棒配准算法；研究二维回归移动平均参数估计,提出模糊盲辩识的快速算法，解决非凸的疑难问题；研究基于凸组合2-范数和1-范数度量的最优估计，研究最优次梯度与共轭下降技术，提出基于适应凸组合2-范数和1-范数度量的图像超分辨率盲重建的快速鲁棒算法，并在理论、计算、应用效果上予以证实。本项目的展开所取得研究成果，对推动图像超分辨率重建技术在理论及实时应用发展具有重要的理论意义和应用价值。