Mamba-Based PP-OCR Enhanced with Super-Resolution for Bus Route Number Recognition

Visually impaired individuals face considerable difficulties in accurately recognizing bus route numbers, which is essential for navigating public transportation systems. While existing approaches typically rely on Optical Character Recognition (OCR) techniques to identify these numbers, their effectiveness is often limited by image degradation caused by blurring and Light Emitting Diode (LED) stroboscopic effects. To address this challenge, we propose a Super-Resolution enhanced Mamba-based OCR framework (SRM-OCR) to improve recognition accuracy for visually impaired users. Specifically, the SRM-OCR adopts a two-stage cascaded design, in which a super-resolution module is employed to enhance image clarity in first stage. In the second stage, a Mamba-based OCR module is integrated to improve the robustness against non-structural noise, leveraging the strong capacity of Mamba to capture long-range dependencies. Additionally, we built a new real-world dataset about bus route number, BusLED-700, which contains diverse low-quality images affected by various distortions. Finally, experimental results demonstrate that the proposed SRM-OCR framework outperforms existing competing methods across various degraded factors, achieving an accuracy of 85.1% and a normalized edit distance (NED) of 0.941 on the BusLED-700 dataset, which confirms its feasibility and reliability in practical applications for visually impaired users.