Phase Retrieval using Untrained Neural Network Priors
Phase Retrieval using Untrained Neural Network Priors
复制标题
DOI：
--
复制
发表时间：
2019-09
期刊：
影响因子：
--
通讯作者：
Gauri Jagatap;C. Hegde
中科院分区：
文献类型：
--
作者：
Gauri Jagatap;C. Hegde
最近，未经训练的深度神经网络作为图像先验被引入用于线性逆成像问题，例如去噪、超分辨率、修复和压缩感知，与稀疏性等手工制作的图像先验相比，具有令人鼓舞的性能提升。此外，与学习的生成先验不同，它们不需要对大型数据集进行任何训练。在本文中，我们考虑压缩相位检索（CPR）的非线性逆问题；这涉及从 n 个仅幅度测量重建 d 维图像信号，其中 n d。为此，我们将图像建模为位于具有固定种子的未经训练的深度生成网络的范围内。然后，我们提出了两种解决 CPR 的方法——梯度下降和投影梯度下降——并且与使用手工设计先验的算法相比，表现出优越的经验性能。
Untrained deep neural networks as image priors have been recently introduced for linear inverse imaging problems such as denoising, super-resolution, inpainting, and compressive sensing, with promising performance gains over hand-crafted image priors such as sparsity. Moreover, unlike learned generative priors they do not require any training over large datasets. In this paper, we consider the non-linear inverse problem of compressive phase retrieval (CPR); this involves reconstructing a d-dimensional image signal from n magnitude-only measurements, where n d. To this end, we model images to lie in the range of an untrained deep generative network with a fixed seed. We then present two approaches for solving CPR — gradient descent, and projected gradient descent — and show superior empirical performance when compared to algorithms that use hand crafted priors.