基于梯度域多测度的湍流空变退化图像复原方法研究

With the development of detector performance, and the detective field of view angle widened, space variant turbulence degradation gradually becomes a bottleneck problem which restricts the development of remote object detection technology. In order to deal with the fast-changing turbulence motion, the randomness of varying point spread functions, computational complexity and overall results are sensitive to computational error, this program aims at turbulence space variant degraded image restoration based on multiple measurements in gradient domain. The main content is as the following. We define statistic measurements in gradient domain and propose point spread function estimation method based on measurements fusion. In order to make blur matrix fits parameters characteristics of turbulence space variant degradation, and preserve details and textural features in restored image, adding parameters constraint term of blur matrix and nonlocal similarity constraint terms as regularization terms, we build turbulence space variant degraded image restoration model. We introduce error matrix to eusure the robustness of model. Then turbulence degraded images are restored. This program will provide new theoretical method and technical support for improving real turbulence degraded space variant image quality, and make the pavement for the real application of turbulence degraded space variant image restoration.

随着探测器性能的发展，探测视场角度的变宽，空变湍流退化逐渐成为严重制约着现在及未来远距离目标探测技术发展的瓶颈问题。由于湍流运动的瞬息万变特性，点扩展函数的随机变化多样性，计算复杂性及整体结果对计算误差的敏感性，本项目开展基于梯度域多测度的湍流空变退化图像复原方法研究。主要内容包括：为了精确地估计点扩展函数，本项目在梯度域定义利于点扩展函数估计的统计测度，提出基于测度融合的点扩展函数估计方法；为了使模糊矩阵更符合湍流空变退化主要参数化特性，并保留恢复图像中的细节和纹理特征，以模糊矩阵的湍流退化参数约束项和对图像施加非局部相似性约束为正则化项，建立湍流空变退化图像复原模型，并引入误差矩阵使空变模型具有鲁棒性，对湍流空变退化图像实现快速优化复原。本研究成果为提高实际湍流空变退化图像的质量提供新的理论方法和技术支撑，为实际湍流退化空变复原的应用奠定扎实基础。

本项目开展了基于梯度域多测度的湍流空变退化图像复原方法研究，发表论文18篇，其中9篇外文期刊，4篇国内期刊，5篇会议。申请发明专利2项，获得2020年自动化协会科技进步奖1项(排名第4)。研究成果主要包括: 首先，分析湍流退化空变图像的梯度域，充分利用梯度域信息，定义了基于方向梯度直方图、梯度、模糊带宽度的测度，并融合梯度域的多测度，获取利于模糊核估计的有效图像内容。其次，为了提升空可变图像复原的速度和获取合理的图像块，对空可变退化图像进行自适应分块，建立基于梯度域多测度的多正则化约束的湍流退化空变复原模型。最后，提出了多正则化约束的湍流退化空变图像的快速鲁棒优化复原方法，多正则化包括：基于显著边缘选择正则化项的湍流退化图像复原、基于分数阶微分算子和自适应加权梯度正则化的湍流退化图像复原、基于卷积神经网络去噪正则化约束的湍流退化图像盲复原、基于去噪约束和深度图像先验正则化的盲图像复原。为提高实际湍流退化空变图像的质量提供新的理论方法和技术支撑，为实际湍流退化空变复原的应用奠定扎实的基础。

内容获取失败，请点击重试