Generative approach for lensless imaging in low-light conditions.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Artificial Intelligence in Imaging

Background:

• Lensless imaging systems offer compact, lightweight alternatives to traditional lens-based optics, suitable for space-constrained applications.
• Low-light conditions in these environments lead to photon starvation and complex noise, degrading image quality.
• Existing reconstruction methods struggle with the noise and information loss inherent in low-light lensless imaging.

Purpose of the Study:

• To develop a robust reconstruction method for high-quality lensless imaging under low-light conditions.
• To address the challenges of noise interference and insufficient photon capture in lensless imaging systems.
• To improve the visual fidelity and diagnostic utility of images acquired in resource-limited environments.

Main Methods:

• A hybrid approach combining a physics-model-driven reconstruction with a data-driven generative model.
• Utilized a learnable Wiener filter for initial noisy reconstruction and a modified conditional generative diffusion module for noise suppression.
• Employed wavelet domain transformation and bidirectional training for efficient and stable image generation.

Main Results:

• Demonstrated significant improvements in image quality and noise reduction compared to conventional methods.
• Successfully reconstructed high-quality images from noisy, low-light measurements in both simulations and real-world experiments.
• Validated the effectiveness of the hybrid model-driven and data-driven approach for challenging imaging scenarios.

Conclusions:

• The proposed method significantly enhances lensless imaging capabilities in low-light environments.
• The dual-perspective reconstruction strategy effectively mitigates noise and improves image clarity.
• This advancement holds promise for expanding the applications of lensless imaging in challenging conditions.