Wavefront coding image reconstruction via physical prior and frequency attention.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Biomedical Engineering

Background:

• Wavefront coding (WFC) is a key technique for extending depth-of-field in imaging systems.
• Traditional WFC involves optical encoding and digital decoding, often facing challenges with noise and artifacts.
• Developing advanced decoding methods is crucial for improving image quality in WFC systems.

Purpose of the Study:

• To propose a novel deep-learning based wavefront decoding method.
• To enhance image reconstruction quality for large depth-of-field imaging systems.
• To validate the method's effectiveness in reducing noise, artifacts, and blur.

Main Methods:

• Applied physical prior information and frequency domain models to wavefront decoding.
• Developed a generative model-based reconstruction method inspired by transformer architecture.
• Introduced three novel modules: PSF attention layer, multi-feature fusion block, and frequency domain self-attention block.
• Utilized a genetic algorithm for phase mask design in a fluorescence microscope for encoding.

Main Results:

• The proposed deep-learning model effectively reconstructs encoded images.
• Experimental results demonstrate significant reduction in noise, artifacts, and blur.
• The method successfully improves image quality for large depth-of-field applications.

Conclusions:

• A deep-learning wavefront decoding model was successfully developed and validated.
• The model offers improved reconstruction image quality for large field-of-view systems with extended depth-of-field.
• This approach shows significant potential for applications such as digital polymerase chain reaction (dPCR) and biological image analysis.