Publication

Abstract :

Scene text detection and recognition (STR) from real-world images has emerged as an important research problem in recent years due to its wide range of applications and challenges. Scene text recognition (STR) comprises the task of localizing the text region in an image captured from the real world and converting the localized text region into actual text. The applications of scene text recognition include image search, language translation, robot navigation, assisting visually impaired people in understanding their surrounding text and industrial applications like various types of industrial automation, Automatic Number Plate Recognition (ANPR), Augmented Reality (AR), and Advanced Driver Assistance System (ADAS) where it helps people to read traffic sign boards, street names, and other textual information on the roads which are helpful for street navigation. STR technologies are widely used in the content retrieval and image indexing fields. There are numerous survey papers addressing scene text detection and recognition, yet there are no major text recognition survey papers that exclusively concentrate on text recognition using segmentation-free approaches. Segmentation-free methods consistently demonstrate superior performance compared to their segmentation-based counterparts. In this paper, we summarize the fundamental problems in the scene text recognition, the various methods that use a segmentation-free approach for text recognition and the analysis of accuracy and failure cases in text recognition.