Edge detection in images using a neuromorphic system

Edge detection is a problem that has been addressed by multiple studies and represents a current challenge in computer vision and graphics computing (GC), where performance is fundamental in applications such as object recognition, image segmentation, and scene understanding. Traditional edge detection methods, such as Sobel and Canny filters, rely on gradient-based techniques, but often suffer from noise sensitivity and computational inefficiency. Neuromorphic systems, inspired by the biological neural mechanisms of the human visual system, have emerged as a promising alternative for edge detection. These systems utilize spiking neural networks (SNNs), oscillatory neural networks (ONNs), and other biologically inspired architectures to achieve high-performance, real-time, and energy-efficient peripheral sensing. Neuromorphic systems provide real-time, low-latency, and energy-efficient edge detection capabilities. Their main difference is that classical image detection is based on frame generation and analysis, where images are processed synchronously; in contrast, neuromorphic vision employs event-based computing, which reduces redundant processing and enables dynamic adaptation to changes in the scene. Inspired by the brain's neural architecture, pulse neural networks process information from the edges through precise pulse timing and synaptic plasticity, resulting in greater robustness in the face of lighting variations, noise, and motion blur. This article explores the principles and methodologies of edge detection using classical systems. It adapts them to new neuromorphic systems, where key components, such as event-based vision sensors, pulse neural networks, and bioinspired algorithms like the Hierarchical and X Model (HMAX) or Neocognitron, are utilized. Several neuromorphic architectures were analyzed, and their performance in real-world edge detection scenarios was evaluated. In addition, neuromorphicbased edge detection was compared with conventional techniques in terms of accuracy, energy efficiency, and processing speed. Experimental results demonstrate that neuromorphic approaches can achieve comparable or superior edge detection performance while consuming significantly less energy and dynamically adapting to changing environments. Applications of neuromorphic edge detection span robotics and autonomous navigation, medical imaging, security surveillance, and augmented reality. The challenges of neuromorphic computing are also discussed, including hardware limitations, training complexity, and the need for optimized learning rules tailored to event-based processing.