High-quality reconstruction method for miniaturized lensless cameras based on imaging model optimization

The multiplexing property of lensless imaging results in the encoded blur image being significantly larger than the effective scene region. The inherent contradiction between blur image truncation and reconstruction quality severely restricts the sensor miniaturization process. To address this issue, this paper proposes a high-quality reconstruction method for miniaturized lensless cameras based on imaging model optimization. This method applies a cropping operator to the scene image, making the reconstruction equations overdetermined, and effectively solves the underdetermined problem encountered when reconstructing from small-sized blur images. An improved Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) is employed for basic scene reconstruction, combined with a U-Net network to further enhance the perceptual quality of the reconstructed image. Experimental results on the simulated dataset OCIFAR100_sim, the captured dataset OCIFAR100, and the public dataset PhlatCam show that, compared with traditional optimization algorithms and mainstream deep learning reconstruction algorithms, the proposed methods can achieve more stable and higher-quality scene reconstruction under small-sized blur image input conditions. This research provides theoretical support and technical pathways for the miniaturization of lensless camera sensors, with significant practical application value.