Dynamic-Step-Size Regulation in Pulse-Coupled Neural Networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Computational Neuroscience

Background:

• Pulse-coupled neural networks (PCNNs) offer unsupervised image segmentation but face challenges in optimal output selection.
• The step size in PCNNs critically affects membrane potential dynamics and thresholding, influencing segmentation outcomes.

Purpose of the Study:

• To develop an adaptive PCNN model for improved unsupervised image segmentation.
• To enhance the control over segmentation granularity and robustness against noise.

Main Methods:

• Proposed a dynamic-step-size mechanism using trigonometric functions to adaptively control segmentation.
• Implemented supervised optimization of a single parameter (ϕ) via Intersection over Union (IoU) maximization.
• Evaluated model performance on image segmentation tasks, including robustness to noise.

Main Results:

• The dynamic-step-size mechanism allows controllable segmentation granularity and increased model adaptability.
• Achieved superior performance compared to existing SPCNN and PCNN methods, with IoU = 0.8863 and Dice = 0.901.
• Demonstrated enhanced robustness under noise (92.1% Dice at σ=0.2) and efficient processing (0.8684 s/image).

Conclusions:

• The proposed dynamic-step-size PCNN model significantly improves unsupervised image segmentation.
• The method offers better adaptability, noise robustness, and reduced tuning complexity for diverse applications.
• This advancement contributes to more effective and reliable image segmentation techniques.