Deep Learning Enhanced Color Super-Resolution Imaging in Metalens-Integrated Camera

Metalenses are ultra-light and ultra-thin imaging elements composed of two-dimensional arranged sub-wavelength nanostructures, which holds considerable promise for the advancement of highly integrated cameras. However, the performance of metalens-integrated cameras is limited by their fixed architectures, particularly in terms of imaging resolution. To address this constraint, we propose a deep learning enhanced super-resolution method for color imaging based on an achromatic metalens. We employ a multi-scale super-resolution convolutional neural network (MSCNN) to train an extensive paired dataset comprising high-resolution and low-resolution colorful images acquired from experiments. Postprocessing with our MSCNN yields a notable improvement in Peak Signal-to- Noise Ratio (PSNR) of approximately 5.8 dB, with Structural Similarity Index (SSIM) exceeding 0.9. Our approach empowers cameras to combine the benefits of high integration with enhanced color imaging performance, unveiling tremendous potential for future groundbreaking imaging technology.