Noise-robust lensless imaging via high-frequency information-guided phase retrieval

Lensless digital in-line holographic microscopy (LDHM) has emerged as a powerful technique for complex wavefront reconstruction. However, the frequency-domain information encoded in diffraction patterns varies with defocus distance due to wave-vector evolution. Conventional phase retrieval algorithms typically neglect these frequency-dependent variations, which limit suppression of noise-induced artifacts and fail to resolve the twin-image problem. In this work, we propose a high-frequency information-guided phase retrieval (HGPR) algorithm that exploits the differential frequency content across multiple diffraction patterns. By using measurements with richer high-frequency components as guidance constraints, the proposed framework enhances reconstruction fidelity by more effectively extracting fine details. Furthermore, we integrate an adaptive weighting scheme into a constrained optimization framework, which leverages measurement diversity to achieve improved spatial resolution. Compared with conventional algorithms, HGPR not only achieves superior resolution enhancement but also requires fewer measurement frames. Both simulations and experiments demonstrate that the proposed method significantly improves reconstruction fidelity, robustness against noise, and spatial resolution.