S5-3 - SNR Enhancement in Event-Based Grayscale Image Reconstruction via Pixel Sensitivity Characterization

Event

23. ITG/GMA-Fachtagung Sensoren und Messsysteme 2026
2026-06-09 - 2026-06-10
Nürnberg

Band

Vorträge

Chapter

Event-Based Vision

Author(s)

A. L. Delucchi, J. Baßler, M. Heizmann - Karlsruhe Institute of Technology, Karlsruhe, P. Bäcker - Fraunhofer IOSB, Karlsruhe

Pages

98 - 104

DOI

10.5162/sensoren2026/S5-3

ISBN

978-3-910600-11-9

Price

free

Abstract

This paper presents a robust method for enhancing the signal-to-noise ratio (SNR) in grayscale images reconstructed via the Ev₂Gray method from event-based vision sensors. Reconstructed images often exhibit significant noise originating from the sensor’s inherent fixed-pattern noise (FPN). We propose a twophase technique involving the empirical extraction of a pixel sensitivity matrix followed by an algorithmic normalization. By adopting a logarithmic sensor response model, the multiplicative noise is transformed into a linear additive problem, allowing for computationally efficient correction. The contributions of this work are two-fold: First, we introduce a practical and straightforward method to quantify the FPN of a sensor without the requirement of laboratory equipment. Second, we demonstrate that using the knowledge of this quantified FPN, we can significantly improve image quality. The proposed dynamic parameter estimation enables the algorithm to adapt to varying conditions, such as temperature and gain fluctuations, without requiring manual recalibration. Experimental results demonstrate a substantial reduction in structural correlation between the FPN and the reconstructed images, decreasing from 0.009, respectively, while maintaining the global mean intensity. This r = 0.257 to approach effectively suppresses persistent sensor-level artifacts across diverse operating scenarios, providing a critical advancement for high-quality semantic and structural scene interpretation in event-based vision. 0.063 and r = 0.138 to − −