DSCCNet for high-quality 4K computer-generated holograms

Area of Science:

• Optics and Photonics
• Computer Vision
• Deep Learning

Background:

• High-quality 3D holographic reconstruction faces challenges in visual clarity and accuracy.
• Existing methods often suffer from low resolution, detail loss, and checkerboard artifacts.

Purpose of the Study:

• To introduce a novel deep learning model, DSCCNet, for phase-only computer-generated holography (CGH).
• To enhance reconstruction precision, improve image resolution, and reduce artifacts in 3D holographic displays.

Main Methods:

• Developed the Depthwise Separable Complex-valued Convolutional Network (DSCCNet) integrating complex-valued and depthwise separable convolutions.
• Incorporated a diffuser to mitigate checkerboard artifacts in defocused regions of 3D CGH.
• Utilized the DIV2K dataset for validation.

Main Results:

• DSCCNet achieved 4K image reconstructions with significantly more intricate details.
• Demonstrated enhanced reconstruction quality for both 2D and 3D layered objects.
• Achieved an average Peak Signal-to-Noise Ratio (PSNR) above 37 dB and an average Structural Similarity Index Measure (SSIM) above 0.95 on 100 DIV2K images.

Conclusions:

• DSCCNet offers an effective solution for high-quality 3D holographic reconstruction.
• The model addresses key limitations of current CGH techniques, improving visual fidelity.
• This advancement is crucial for the growing demand in advanced holographic imaging applications.