基于单帧图像的复杂光学成像条件下图像复原方法

Image restoration method based on single-frame image under complex optical imaging conditions

大气层、临界层等光学成像环境复杂,如何准确求解点扩散函数并抑制噪声是图像复原的两个重点。针对目前复原算法存在先验信息利用不足和抗噪性能较差等问题,提出基于单帧图像的复原方法。首先建立多尺度多方向形态滤波器,引入噪声特征图像自适应选择结构元素,克服传统方法固定结构元素无法改变的缺点;使用小波变换将小波的多分辨率特性与图像复原相结合,加快运算速度并进一步抑制残余噪声;提出基于分形维数动态求解支持域与正则化参数,改善以往固定支持域的缺点,提升图像复原的准确性。大量实验结果表明,本算法仅使用单帧模糊图像即可有效复原,可适用于多种类型数据集,满足实时复原的场景。

The optical imaging environment,such as the atmosphere and critical layer,is complex.Accurately solving the point spread function and suppressing noise are the two main focuses of image restoration.In order to address the problems of insufficient use of priori information and poor noise immunity in current restoration algorithms,a restoration method is proposed based on single-frame image.Firstly,a multi-scale multi-directional morphological filter is established,while the structure elements are adaptively selected using the noisy feature image to overcome the shortcomings of the traditional method that the fixed structure elements cannot be changed.The wavelet transform is used to combine the multi-resolution properties of wavelets with image restoration to speed up operations and further suppress residual noise.The support domain and regularization parameters are dynamically solved based on the fractal dimension to improve the shortcomings of the previous fixed support domain and enhance the accuracy of image restoration.Extensive experimental results show that the algorithm can effectively reconstruct blurred image using only single-frame image,and can be applied to many types of data sets to meet the scenario of real-time restoration.