Optoelectronic thresholding module for winner-take-all operations in optical neural networks

Area of Science:

• Optoelectronics
• Optical Computing
• Image Processing

Background:

• Optical thresholding is crucial for optical neural networks and image processing.
• Existing architectures may require additional light sources or disrupt beam paths.

Purpose of the Study:

• To propose a simple optoelectronic architecture for optical thresholding.
• To demonstrate its suitability for optical neural network (ONN) and image processing applications.

Main Methods:

• The proposed module integrates liquid-crystal television (LCTV), a CCD camera, and beam-sampling optics.
• It operates without an external light source, ensuring translation invariance.
• The design preserves the beam propagation path.

Main Results:

• Experimental validation was performed using various images and correlation patterns.
• The system demonstrated effective optical thresholding capabilities.
• An architecture for an automatic winner-take-all extractor was also presented.

Conclusions:

• The developed optoelectronic architecture offers a practical solution for optical thresholding.
• It is well-suited for advancing optical neural network and image processing technologies.
• The design's simplicity and efficiency make it a valuable contribution to the field.