Adaptive Guided Filtering and Spectral-Entropy-Based Non-Uniformity Correction for High-Resolution Infrared Line-Scan Images

TL;DR

This paper introduces a new lightweight method to correct stripe noise in high-resolution infrared images, improving quality for real-time applications.

Contribution

A novel single-frame, reference-free non-uniformity correction method with adaptive filtering and spectral-entropy-based noise suppression.

Findings

The method outperforms six state-of-the-art algorithms in simulated and real datasets.

It achieves the fastest runtime while maintaining high performance across multiple metrics.

Abstract

Stripe noise along the scanning direction significantly degrades the quality of high-resolution infrared line-scan images and impairs downstream tasks such as target detection and radiometric analysis. This paper presents a lightweight, single-frame, reference-free non-uniformity correction (NUC) method tailored for such images. The proposed approach enhances the directionality of stripe noise by projecting the 2D image into a 1D row-mean signal, followed by adaptive guided filtering driven by local median absolute deviation (MAD) to ensure spatial adaptivity and structure preservation. A spectral-entropy-constrained frequency-domain masking strategy is further introduced to suppress periodic and non-periodic interference. Extensive experiments on simulated and real datasets demonstrate that the method consistently outperforms six state-of-the-art algorithms across multiple metrics while maintaining the fastest runtime. The proposed method is highly suitable for real-time deployment in airborne, satellite-based, and embedded infrared imaging systems. It provides a robust and interpretable framework for future infrared enhancement tasks.

Figures