End-to-end optical-algorithm co-design for achromatic extended depth-of-field imaging

Abstract

This paper presents an end-to-end optical-algorithm co-design framework for achromatic extended depth-of-field imaging, targeting the trade-off between spatial resolution, depth of field, and chromatic degradation in high-resolution imaging systems. The proposed framework formulates the phase plate in a Fourier imaging system as a differentiable optical encoding layer and parameterizes its height map with Zernike polynomials, enabling low-dimensional and structured control of complex wavefront aberrations. A regularized computational decoder reconstructs the image, while the optical parameters and image-processing parameters are jointly updated under a unified loss function. Experiments on the BSDS100 and Urban100 datasets show that the proposed method produces a more compact point spread function, suppresses defocus blur, chromatic artifacts, and noise amplification, and achieves higher reconstruction PSNR in both natural texture and urban high-frequency scenes. Robustness analysis further indicates that appropriate sparse regularization stabilizes training and improves the preservation of high-frequency details.