A luminance coding strategy for depth adaptive multi-plane holography

Computer-generated holography (CGH) combined with deep learning shows great potential in three-dimensional display. However, most existing methods are limited to a fixed reconstruction distance and thus lack the flexibility required for dynamic holographic display. In this paper, we present a luminance-coded holographic synthesizer (LCHS) that performs distance adaptive multi-plane hologram generation in a single network inference. The proposed method encodes depth information using artificially constructed luminance differences between paired input images, enabling the network to adaptively generate holograms matched to the target reconstruction distances. On this basis, a spatial depth encoder based on local luminance variations is further constructed to extract and utilize local adaptive depth features, thereby supporting high-quality reconstruction of multi-plane scenes. Moreover, by combining the relative luminance based depth representation with diffraction propagation at the prescribed reconstruction distances, the framework can be transferred to previously unseen depth intervals without retraining while maintaining relatively stable reconstruction performance. Experimental results show that the framework maintains reasonably stable reconstruction quality at testing distances up to 50 mm beyond the training range, demonstrating its transferability to previously unseen depth ranges.