Deep learning-based edge enhancement of interferenceless coded aperture correlation holography without a point spread...

Area of Science:

• Optics
• Computer Vision
• Holography

Background:

• Interferenceless coded aperture correlation holography (I-COACH) faces challenges in achieving high-quality edge-enhanced results due to cross-correlation suppression of modulation characteristics.
• Conventional methods often require complex point spread hologram recordings and iterative or nonlinear algorithms for edge enhancement.

Purpose of the Study:

• To develop a deep learning-based method for high-quality edge-enhanced reconstruction in I-COACH.
• To overcome the limitations of existing methods in suppressing modulation characteristics and achieving superior edge enhancement.

Main Methods:

• A U-net architecture was employed to directly establish a mapping between I-COACH object holograms and edge-enhanced objects.
• The proposed method bypasses the need for recording system point spread holograms.
• Testing involved holograms from various phase masks, different initial randomness, and occlusion interference.

Main Results:

• The deep learning method achieved high-quality edge-enhanced reconstruction, outperforming complex iterative and nonlinear algorithms.
• The network demonstrated strong robustness, successfully distinguishing object features and providing acceptable edge enhancement even with 90% hologram occlusion.
• Applicability was shown through edge-enhanced imaging of 3D objects at different depths.

Conclusions:

• The proposed deep learning approach offers a promising strategy for edge enhancement in 3D incoherent imaging.
• This method expands applications in pattern recognition and edge detection by providing robust and high-quality edge-enhanced imaging.
• The U-net based reconstruction effectively addresses I-COACH limitations, enabling advanced imaging capabilities.