Diffractive deep neural network-based depth-of-field expansion without image restoration

Area of Science:

• Optical engineering
• Computational imaging
• Artificial intelligence in optics

Background:

• Lens-based display systems face depth of field (DOF) limitations, causing image blur and distortion.
• Traditional diffractive optical devices (DOE) have challenges in extending DOF effectively.
• Image quality degradation and the need for post-processing are common issues with existing DOF extension methods.

Purpose of the Study:

• To propose and validate a novel depth of field extension method for lens-based display systems.
• To overcome the inherent DOF limitations of lenses using advanced computational techniques.
• To achieve depth-invariant imaging with high fidelity across an extended DOF range.

Main Methods:

• Development of a diffractive deep neural network (DDNN) to replace conventional DOEs for DOF extension.
• Utilizing the Adam algorithm for optimizing the phase distribution within the DDNN.
• Focusing on achieving a depth-invariant and concentrated point spread function (PSF) across the entire DOF.

Main Results:

• The proposed DDNN method successfully extends the depth of field in lens-based display systems.
• The method achieves a depth-invariant and concentrated PSF distribution throughout the extended DOF.
• Demonstrated superior performance compared to existing methods, maintaining imaging quality.

Conclusions:

• The diffractive deep neural network offers a significant advancement in extending the depth of field for display systems.
• This method eliminates the need for post-image restoration, reducing integration complexity and time costs.
• The approach presents a promising solution for high-quality, depth-invariant imaging in optical display technologies.