IFR: Iterative Fusion Based Recognizer for Low Quality Scene Text Recognition

Although recent works based on deep learning have made progress in improving recognition accuracy scene text recognition, how to handle low-quality images end-to-end networks remains a research challenge. In this paper, we propose an Iterative Fusion Recognizer (IFR) for low quality taking advantage of refined input and robust feature representation. IFR contains two branches which focus image recovery respectively. We utilize iterative collaboration between branches, can effectively alleviate the impact input. A fusion module is proposed strengthen representation where features from are Fused with Restoration branch, referred as RRF. Without changing network structure, extensive quantitative qualitative experimental results show that method significantly outperforms baseline methods boosting benchmark datasets resolution TextZoom dataset.