湍流退化图像相位恢复算法研究

Restoration of turbulence-degraded images based on phase retrieval algorithm

为克服气动光学效应对目标图像的影响,把相位恢复算法与气动光学效应机理研究结合起来,用于湍流退化图像的恢复。该算法是通过目标图像的傅里叶变换幅值来恢复目标图像,或等价地,恢复傅里叶变换相位。讨论了两类相位复原算法——迭代傅里叶变换(IFT)和解相关算法。对现有的解相关算法作了改进,采用共轭梯度法解高斯-牛顿方程,可有效提高算法的收敛速度。IFT算法不能保证迭代过程总能收敛到正确解,有时会出现停滞现象,将IFT和解相关算法组合起来,可以克服IFT算法的停滞现象,提高正确收敛率。给出了在信噪比为20 dB情况下的湍流退化仿真图像恢复的实例,目标图像能较清晰地恢复出来。实验结果表明该算法具有较好的稳定性和抗噪声能力。

To suppress the aero-optic effect on object images,the phase retrieval algorithm is applied together with aero-optic mechanism to restore turbulence-degraded images.The principle of this algorithm is to reconstruct object images from their Fourier transform magnitude and is equivalent to the principle of the reconstruction of the Fourier phase.Two methods for phase retrieval,namely the iterative Fourier transform （IFT） algorithm and a new de-autocorrelation algorithm, are discussed.The current de-autocor- relation algorithm is improved by using the conjugate gradient（CG）method to solve the Gauss-Newton equation in order to accelerate convergence.The reconstruction done by the IFT algorithm does not always converge to correct results,but combination of the IFT algorithm and the de-autocorrelation algorithm will reduce the stagnation of the IFT algorithm and increase the effectiveness of phase retrieval.Expermental examples of restoration of a simulated turbulence-degraded image with white noise at signal-to-noise ratio （SNR）of 20 dB are given. The experimental results show that the recommended method is an effective method and less sensitive to noise.