SNNF: An SNN-based Near-Sensor Noise Filter for Dynamic Vision Sensors

2026-05-05 · Source: cs.NE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Internet of Things (IoT) & Connected Devices · Depth: Expert, extended

Summary

SNNF is a near-sensor noise filter designed for Dynamic Vision Sensors (DVS) to mitigate spurious Background Activity (BA) noise, which degrades output and increases computational overhead in edge applications. It integrates a compact Event-Based Binary Image (EBBI) representation, a parallel memory architecture, and a single-layer Spiking Neural Network (SNN) classifier. The SNNF achieves an Area Under the Curve (AUC) of 0.89 on standard DVS datasets, effectively distinguishing signal events from noise. Its binary-array-based EBBI eliminates timestamp dependency, significantly reducing memory footprint, while the SNN's spike-based computation replaces power-hungry multipliers with accumulation logic, minimizing inter-neuron data width. FPGA implementations show SNNF reduces memory and logic resources to approximately 11% and 40%, respectively, of state-of-the-art filters, achieving a throughput of 29 Mega events per second (Meps). A 65 nm CMOS ASIC implementation achieves 44.4 Meps with an area and power consumption of only ~13% and <5% of corresponding ANN-based designs, demonstrating an excellent balance between filtering accuracy and hardware efficiency for resource-constrained, near-sensor deployment.

Key takeaway

For Computer Vision Engineers developing low-power, real-time DVS applications, SNNF offers a compelling solution for near-sensor noise filtering. Its demonstrated superior hardware efficiency, including significantly reduced memory (11% of alternatives) and power consumption (<5% of ANN-based designs), combined with high accuracy (AUC 0.89) and throughput (44.4 Meps on ASIC), means you can deploy robust denoising directly on edge devices without compromising performance or resource constraints. Consider integrating SNNF's architectural principles to optimize your next-generation neuromorphic vision systems.

Key insights

SNNF efficiently filters DVS noise using a compact binary representation and a hardware-optimized Spiking Neural Network.

Principles

• Spatiotemporal correlation distinguishes signal from noise.
• Binary event representation reduces memory footprint.
• SNNs offer energy efficiency via spike-based computation.

Method

SNNF converts DVS events into EBBI pairs, extracts local patches, and feeds them to a single-layer FCSNN for classification, utilizing a parallel memory architecture for high-throughput patch extraction.

In practice

• Use constant-time EBBI generation with T_e=25ms for optimal performance.
• Employ a 1-layer FCSNN with ~30 hidden neurons for efficiency.
• Distribute EBBI data across N_mem=n memory banks for parallel access.

Topics

• Dynamic Vision Sensors
• Spiking Neural Networks
• Event-Based Binary Image
• Near-Sensor Noise Filtering
• Hardware Efficiency

Best for: Computer Vision Engineer, Research Scientist, AI Hardware Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• FS-DVS: A Frequency-Selective Dynamic Visual Sensing Paradigm for Enhancing Information Completeness — FS-DVS uses a learnable pre-trigger spatial filter to mimic RGCs, enhancing event camera data completeness and noise resilience.
• The Third Challenge on Image Denoising at NTIRE 2026: Methods and Results — The NTIRE 2026 challenge benchmarks unconstrained neural architectures for high-noise image denoising, prioritizing PSNR performance.
• Energy-Aware Spike Budgeting for Continual Learning in Spiking Neural Networks for Neuromorphic Vision — An energy-aware spike budgeting framework improves SNN continual learning by adaptively balancing accuracy and energy across vision modalities.
• Efficient Sensor Fusion for Gesture Recognition on Resource-Constrained Devices — Sensor fusion of low-resolution ToF and IR enables efficient, private gesture recognition on resource-constrained devices.
• Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision — The framework re-tokenizes high-resolution event streams into compressed, informative "neural events" using discrete learnable codes.
• Plug-and-Play image restoration with Stochastic deNOising REgularization — SNORE improves PnP image restoration by applying denoisers to appropriately noisy images, enhancing performance and theoretical guarantees.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.NE updates on arXiv.org.