Lensless holographic dynamic projection system based on weakly supervised learning

Diffraction-based lensless projection system offers the advantages of a compact size with high flexibility. However, existing iterative methods of computation for computer-generated holograms (CGHs) in projection systems are not mature enough. Furthermore, slow CGH generation process is not suitable in dynamic display pattern. Here, a lensless projection system with large field of white light without chromatic aberration is proposed, enabling rapid hologram generation and white light display utilizing a single hologram under trichromatic illumination. The core innovation lies in leveraging differences in reconstructed images from holograms acquired under different wavelength conditions to constrain network training. Additionally, a lightweight hologram generation network is designed for hologram synthesis. This network is trained in a weakly supervised manner, utilizing the Gerchber-Saxton (GS) iterative algorithm based on the scaled angular spectrum method (SASM–GS) to generate reference holograms. The utilization of a weakly supervised approach enhances the efficiency of network training and overcomes the limitations imposed by label-based training. Simulation results demonstrate that the proposed method can rapidly generate holograms with low chromatic aberration, achieving a hologram generation speed of 40 frames per second while maintaining a peak signal-to-noise ratio of 35 dB for reconstructed images. A lensless projection system based on digital micromirror device (DMD) is constructed, and optical experiments confirm that the proposed method enables color-consistent white light display using a single hologram, making it applicable to low-cost, small-scale projection systems and other domains.