Unified Thermal Image Enhancement for Object Detection and Visual Clarity

Thermal infrared imaging systems are increasingly important in autonomous driving, medical diagnostics, and solar panel inspection applications. However, the effectiveness of thermal images is often limited by low contrast, indistinct details, uneven illumination, and various artifacts that hinder human interpretation and automated analysis. Conventional and deep learning-based enhancement methods offer partial solutions but frequently struggle to generalize across diverse conditions or produce consistently high-quality results. This paper proposes a unified thermal image enhancement pipeline that combines decomposition-based image processing, robust illumination balancing, and analytically designed pseudo-coloring schemes to enhance visual clarity, preserve structural details, and deliver interpretable colorized outputs suited for downstream tasks. Our methodology consists of a three-stage pipeline: (1) decomposition to separate fine details from the base structure, (2) adaptive contrast enhancement using the specialized module, and (3) pseudocoloring that maps thermal intensities to perceptually meaningful color spaces. The pipeline enhances visual clarity, preserves structural details, and delivers interpretable colorized outputs suited for downstream tasks. Extensive experimentation on multiple real-world datasets, including autonomous vehicles, demonstrates that our method significantly improves visual quality and object detection performance. Furthermore, evaluation of the proposed enhancement and coloring techniques using the YOLOv7 object detector shows that our approach yields up to 2.6% higher mAP0.5 compared to models trained on original thermal images, highlighting its effectiveness in improving object detection accuracy.