PhoCoLens: Photorealistic and Consistent Reconstruction in Lensless Imaging

Xin Cai,Zhiyuan You,Hailong Zhang,Wentao Liu,Jinwei Gu,Tianfan Xue
2024-10-07
Abstract:Lensless cameras offer significant advantages in size, weight, and cost compared to traditional lens-based systems. Without a focusing lens, lensless cameras rely on computational algorithms to recover the scenes from multiplexed measurements. However, current algorithms struggle with inaccurate forward imaging models and insufficient priors to reconstruct high-quality images. To overcome these limitations, we introduce a novel two-stage approach for consistent and photorealistic lensless image reconstruction. The first stage of our approach ensures data consistency by focusing on accurately reconstructing the low-frequency content with a spatially varying deconvolution method that adjusts to changes in the Point Spread Function (PSF) across the camera's field of view. The second stage enhances photorealism by incorporating a generative prior from pre-trained diffusion models. By conditioning on the low-frequency content retrieved in the first stage, the diffusion model effectively reconstructs the high-frequency details that are typically lost in the lensless imaging process, while also maintaining image fidelity. Our method achieves a superior balance between data fidelity and visual quality compared to existing methods, as demonstrated with two popular lensless systems, PhlatCam and DiffuserCam. Project website: <a class="link-external link-https" href="https://phocolens.github.io/" rel="external noopener nofollow">this https URL</a>.
Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning
What problem does this paper attempt to address?
The paper attempts to address the challenge of achieving high-quality image reconstruction in lensless imaging systems. Specifically, while lensless cameras have significant advantages in terms of volume, weight, and cost, they lack focusing lenses, which leads to their reliance on computational algorithms to recover scenes from complex diffraction patterns. Existing algorithms face two main issues when recovering high-quality images: first, the forward imaging model is inaccurate, and second, there is insufficient prior information to reconstruct high-fidelity images. To address these issues, the paper proposes a new two-stage method called PhoCoLens for consistent and realistic lensless image reconstruction: 1. **First Stage**: Ensure data consistency. Accurately reconstruct low-frequency content through a Spatially-Varying Deconvolution method, which can adapt to changes in the point spread function (PSF) across the camera's field of view. 2. **Second Stage**: Enhance image realism. Utilize a pre-trained diffusion model to generate priors, combining the low-frequency content recovered in the first stage to effectively reconstruct high-frequency details while maintaining image fidelity. Through this method, PhoCoLens achieves a better balance between data fidelity and visual quality, outperforming existing methods. The paper validates the effectiveness of this method through experiments on two popular lensless systems (PhlatCam and DiffuserCam).