Dynamical System Approach for Edge Detection Using Coupled FitzHugh-Nagumo Neurons

Area of Science:

• Computational neuroscience
• Analog circuit design
• Image processing

Background:

• Biologically motivated models are explored for analog circuit design, particularly for image processing tasks like edge detection.
• Existing reaction-diffusion models, while effective, face scalability issues due to a one-to-one pixel-to-neuron mapping.

Purpose of the Study:

• To develop a simplified reaction-diffusion model for efficient edge detection in analog very large scale integration (VLSI) CMOS circuits.
• To overcome the network size limitations of previous biologically inspired image processing models.

Main Methods:

• A simplified reaction-diffusion model was developed in three steps, incorporating continuous Lyapunov exponents for threshold analysis.
• Anisotropic diffusion was introduced, controlled by image grayscale gradients.
• Coupling terms between adjacent neuron membrane potentials were eliminated to simplify the model.

Main Results:

• The simplified model successfully detected edges in both artificial and real image datasets.
• The technique demonstrated performance on par with or superior to existing methods.
• Crucially, the simplification avoided an increase in network size.

Conclusions:

• The developed simplified reaction-diffusion model offers an efficient approach for edge detection in analog circuits.
• This method provides a scalable solution for biologically inspired image processing in VLSI CMOS technology.
• The findings pave the way for more compact and powerful neuromorphic computing systems.