Multi-oriented Text Detection In Natural Images Using Deep Learning
Rana Afreen, Rakesh Huded · 2022 IEEE 2nd Mysore Sub Section International Conference (MysuruCon) · 2022
The exigent demands of text detection have led to the development of numerous deep learning models. Scene text is the text that is incorporated into natural scenes. The precise and explicit spotting and reading of scene text are crucial in countless computer vision applications. This paper presents a novel approach for multi-oriented scene text detection by introducing a deep learning model that integrates CRAFT (Character Region Awareness for Text Detection), text detector and transformer OCR, text recognizer. The CRAFT detector detects the text in the scene text character-wise by producing region and affinity scores. The output of this detector is then fed to the transformer OCR which then gives the recognized text. The results are evaluated using ICDAR 2015 dataset. Hence, this paper fuses two robust mechanisms for extracting textual information from an image, which makes it more accurate than the existing methods.