A Survey on Scene Text Recognition in Natural Images

Text recognition from natural images is a prominent area in computer vision research, given its significance in a multitude of computer vision applications. Scene text detection in natural images has achieved significant advancements in the last years, thanks to the emergence and advancement of deep learning, effectively, many methods have shown remarkable performance. This paper gives a survey on representative scene text recognition methods and highlights the two primary approaches employed in recent works in text recognition from natural images: Attention mechanism and Connectionist Temporal Classification (CTC). This survey also includes a comparative study of these methods based on their performance on benchmark datasets.