Real-World License Plate Image Super-Resolution via Domain-specific Degradation Modeling

License plate (LP) recognition systems often struggle to accurately recognize images in complex environments. Recent studies have attempted to improve recognition accuracy by utilizing super-resolution (SR) technology. However, these approaches often fall short in terms of generalization performance, as they rarely consider the various degradations present in real-world images. In this paper, we propose to generate realistic degraded LP images by applying a degradation model on a high-resolution LP dataset, which can cover a wide range of the degradation variations of real-world LP images flexibly. The SR model trained with simulated degraded images has better generalization and robustness on real-world LP images. Experimental evaluations conducted on LP recognition benchmark datasets demonstrate that the proposed method not only produces visually superior results but also effectively improves recognition accuracy.