基于模糊与畸变可区分性矩特征构造的水下目标湍流退化图像复原方法

With the continuous application of imaging detector in military and civilian fields, the geometric distortion and blurring of the target image caused by water turbulence have become one of the bottlenecks in the development of underwater imaging detection technology. As geometric distortion and blur are two different degradation modes, they interfere with each other and are difficult to express in the process of image restoration. To solution this problem, we research the underwater target turbulence degraded image restoration method which based on blur and distortion differentiable moments feature. The main content includes: first of all, the blur and distortion differentiable moments feature is constructed, and then the entire image restoration framework is set up in three stages. 1）In the frame selection stage, a frame selection algorithm which can independently evaluate the sharpness of degraded image and distortion between frames is proposed. 2）In distortion correction stage, we improve the precision of estimated distortion field between turbulence degraded frames by adding the moment feature constraint in the classical optical flow model and use robust principal component analysis method to suppress the interference and data redundancy which caused by the local abnormal distortion. 3）In the deblurring stage, the image contour intensity which defined by the moments feature is used to control the edge extraction of the degraded image, and the traditional two-stage deblur method is modified to obtain the final recovery result. The results of this study will provide a new theoretical method and technical support for improving the quality of actual underwater turbulent degradation images.

随着成像探测器在军事和民用领域的不断应用，由水湍流引起的目标图像几何畸变和模糊逐渐成为制约水下目标成像探测技术发展的瓶颈。由于几何畸变与模糊是两种不同的退化模式，在图像复原过程中，它们相互干扰，难以表达。针对这一问题，本项目研究基于模糊与畸变可区分性矩特征构造的水下目标湍流退化图像复原方法，先构造能够独立刻画几何畸变与模糊的矩特征，在此基础上分三个阶段搭建整个复原框架。1）在帧选择阶段，提出一种能够独立评价退化图像清晰度与各帧畸变差异的帧选择算法。2）在畸变校正阶段，通过增加具有模糊不变性的矩特征约束项，提高各退化帧之间畸变场估计精度，采用鲁棒性主成分分析法，抑制帧间局部异常畸变带来的干扰及数据的冗余。3）在去模糊阶段，利用矩特征定义的轮廓强度来控制退化图像边缘提取，对传统的两阶段去模糊方法进行改进，得到最终复原结果。本项目研究成果为提高实际水湍流退化图像的质量提供新的理论方法和技术支撑。

本项目研究了基于模糊与畸变可区分性特征构造的湍流退化目标图像复原方法。由湍流效应引起的目标图像几何畸变和模糊严重影响了光学成像系统的性能，阻碍了水下目标成像探测技术的发展，针对湍流效应引起的目标图像质量退化问题，本项目进行了理论创新和实验验证研究，按计划完成了项目研究任务，主要研究成果为：研究和提出了基于模糊与几何畸变可区分性特征表示的湍流退化图像复原方法；提出了基于视觉灵敏性滤波的局部最大差异先验的盲图像复原方法；提出了基于核偏差估计的模糊图像快速复原算法；为了加快复原模型中矩特征的计算速度，研究了基于一阶矩的快速算法；研究了基于图像分层的水下非均匀光照图像复原方法，对复原模型进行优化设计。本项目解决了几何畸变与模糊在湍流退化目标图像复原过程中相互干扰关键问题，开发了水下目标退化图像复原算法软件演示平台，为提高实际湍流退化图像的质量提供新的理论方法和技术支撑。