【作者】 张志怡； 杨义校； 陶然；

【Author】 ZHANG Zhiyi;YANG Yixiao;TAO Ran;School of Information and Electronics, Beijing Institute of Technology;Beijing Key Laboratory of Fractional Signals and Systems;

【机构】 北京理工大学信息与电子学院； 分数域信号与系统北京市重点实验室；

【摘要】 相位恢复问题存在于众多领域,旨在从信号复变换的幅度信息中重建原始信号。在实际应用中,相位恢复算法的计算速度和鲁棒性是两个非常重要的性能指标,因此为优化恢复性能,提高图像重构质量,提出了一种加速收敛的即插即用相位恢复算法。将相位恢复问题表示为一个正则优化问题,利用半二次方分裂法(HQS)将此问题分解为两个独立的子问题;结合LBFGS方法和精确线搜索步长算法,利用拟牛顿方法解决数据保真度子问题;引入DnCNN去噪网络解决正则化子问题。最后在不同的噪声条件和观测模型下开展了对照实验,实验结果表明所提算法实现了快速收敛,有效降低了算法迭代次数,并且对噪声更鲁棒,重构质量更好。更多还原

【Abstract】 The phase retrieval problem exists in numerous fields and aims to reconstruct the original signal from the amplitude information of the complex transform. In order to optimize the recovery performance and improve the quality of image reconstruction, an accelerated plug-and-play phase retrieval algorithm is proposed. Firstly, the phase retrieval problem is represented as a regular optimization problem, and the problem is decomposed into two independent subproblems using the half-quadratic splitting(HQS). Secondly, the data fidelity term is solved using the Quasi-Newton method in combination with the LBFGS method and exact line search stepsize. Then, a framework is proposed to work out the regularization term, which inserts DnCNN denoising network. Finally, controlled experiments are carried out under different noise conditions and measurements. The experimental results show that the proposed algorithm achieves fast convergence and effectively reduces the iteration number. Moreover, it is more robust to noise and has excellent reconstruction quality.更多还原