[3DSAp2-3] Manifold Learning in the Hologram Domain: a Pipeline for Compressing and Reconstructing Complex Holograms
Manifold learning, Computer-generated hologram, Deep neural network, Autoencoder, Hologram domain
We introduce a pipeline for manifold learning in the hologram domain. A traditional autoencoder architecture is employed to compress and reconstruct complex holograms. The pipeline utilizes MNIST data and demonstrates consistent shape reconstruction. The latent vector representation of the hologram exhibits well-separated manifolds, successfully capturing similarities between numbers.