CFConv: complex-valued Fourier convolutional layers for artifact suppression in cylindrical holographic reconstruction

Cylindrical holography has received increasing attention in recent years as a promising technology for achieving ideal 360° full-view displays. However, due to the presence of different obliquity factors in asymmetric diffraction models, the propagation process exhibits inherent asymmetry, posing significant challenges to existing optimization methods. This asymmetry leads to high-frequency information loss and energy attenuation in the frequency domain, which can introduce artifacts near object boundaries and severely degrade the quality of reconstructed images. To address this issue, we propose complex-valued Fourier convolutional layers, specifically designed to compensate for frequency-domain imbalance introduced during the diffraction process. During the phase prediction stage, the proposed complex Fourier convolutional layers effectively preserve high-frequency information around object contours, thereby enhancing the fidelity and sharpness of the reconstructed images. Numerical experiments demonstrate that the proposed method effectively suppresses reconstruction artifacts and improves image quality in cylindrical computer-generated holography tasks.