GS-RNN: a phase-only hologram generation model based on the fusion of the Gerchberg-Saxton algorithm and a neural network

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Traditional Gerchberg-Saxton (GS) algorithm for hologram generation is iterative and can be computationally intensive.
• Phase-only holograms (POHs) are crucial for various optical applications, demanding efficient generation methods.

Purpose of the Study:

• To propose a novel computer-generated hologram (CGH) iterative generation model named GS-RNN.
• To enhance the quality and reduce the iteration count for generating phase-only holograms (POHs).

Main Methods:

• Integration of a recurrent neural network (RNN) architecture with an improved residual complementary neural network.
• Combining the neural network with the traditional Gerchberg-Saxton (GS) algorithm.
• Training the model with sample data to incorporate trainable parameters into the GS algorithm.

Main Results:

• GS-RNN demonstrated iterative computation efficiency three times greater than the traditional GS algorithm.
• The model maintains high performance without retraining, even when physical parameters for POH generation are altered.
• Optical experimental results validated the simulation findings, confirming the model's effectiveness.

Conclusions:

• The proposed GS-RNN model offers a significant advancement in CGH generation, particularly for POHs.
• The model's efficiency, robustness, and accuracy are validated through simulations and optical experiments.
• GS-RNN presents a promising approach for high-quality, efficient hologram generation in various applications.