Text Enhancement Network for Complex Multi-line Scene Text Image Super-resolution

Abstract

Scene text image super-resolution aims to enhance low-resolution, spatially distorted text images by reconstructing high-resolution counterparts with clear textual structure and superior visual fidelity. Nevertheless, the prevalence of multi-line text in real-world scenarios presents a significant challenge to existing text image super-resolution methods. Current approaches predominantly focus on enhancing single-line text images, and their model architectures are inherently constrained when confronted with the complexity of multi-line text. This architectural limitation severely impedes their direct application to more realistic, multi-line text scenarios commonly encountered in natural images. To address these limitations, we propose a Text Enhancement Network for Complex Multi-line Scene Text Image Super-resolution (ML-TSR), a novel framework specifically engineered for multi-line text image super-resolution. ML-TSR employs a two-stage approach: initially reconstructing high-resolution text images, followed by a refinement phase leveraging text-related loss derived from text detection and optimization modules. This process yields enhanced text recovery outcomes. Extensive evaluations on the Real-CE benchmark dataset demonstrate the superior performance of ML-TSR in text restoration quality compared to the state-of-the-art baseline methods.