基于随机点扩散函数的多帧湍流退化图像自适应复原方法

针对湍流退化图像随机性的问题,提出了一种基于随机点扩散函数的多帧湍流退化图像自适应复原方法。首先介绍了随机点扩散函数的图像退化模型,并分析了点扩散函数随机性对图像复原造成的影响,建立了基于随机点扩散函数的多帧图像退化模型。在此基础上,建立了基于多帧退化图像的全变分复原模型,利用前向后向算子分裂法对模型进行求解,提高了算法的运算效率。然后,提出了一种新的自适应正则化参数选取方法,该方法利用全变分复原模型的目标函数计算正则化参数,当正则化参数收敛时,复原图像的峰值信噪比达到最大值,因此利用目标函数的相对差值作为自适应算法迭代终止的条件,可以获得最佳复原效果。最后通过实验分析,算法中退化图像的帧数应不大于10帧。实验结果表明:当取10帧退化图像时,AFBS算法运算时间与单帧的FBS算法相当,信噪比增益为1.4 dB。本文算法对图像噪声有明显的抑制作用,对湍流退化图像可以获得较好的复原效果。

Adaptive restoration method of multi-frame turbulence-degraded images based on stochastic point spread function

As the turbulence-degraded images are stochastic, an adaptive restoration approach of multi-frame turbulence-degraded images was proposed based on stochastic Point Spread Function(PSF). Firstly, an image degradation model of stochastic PSF was introduced, and the influence of the model on the image restoration was analyzed. The degradation model of multi-frame images based on stochastic PSF was established. On this basis, the TV restoration model based on multi-frame images was established. In order to improve the computational efficiency of the algorithm, the model was solved by Forward-Backward Splitting(FBS) operator. Then a new adaptive selection method of regularization parameter was proposed. When the regularization parameter which was calculated by the objective function of the TV model was convergent, the Peak Signal-to-Noise Ratio(PSNR) of restoration image reached the maximum value. In order to get the best restoration effect, the relative error of the objective function was used as the iterative termination condition of the adaptive algorithm. Finally, the number of degraded images should be no more than 10 frames through the experimental analysis. Experimental results show that the ISNR of the AFBS algorithm has increased 1.4 dB more than the FBS algorithm based on single frame while the computing time is comparative when the number of degraded images was 10 frames. The proposed algorithm has an obvious inhibition on the noises, and it can obtain a better restoration effect on turbulence-degraded images.