Extraction of printed information from different certifiable pictures in machine readable format is treated as tricky chore in the domain of computer vision. This has gained plenty of consideration due to its extensive applications in industries and our daily life. Recently developed deep learning based techniques for detection of text have obtained encouraging results with respect to different standard datasets. But, they typically fall behind when exposed to tricky situation. Still there is a contest between speed and accuracy for text detection. Traditionally for text detection mostly the two stage detector techniques like Faster-RCNN, Fast-RCNN and R-CNN were explored. In this work, we put forward a straightforward pipeline for detecting text in common scene pictures. Here we used recent state of art one stage object detection framework known as YOLOv4 (You Only Look Once version 4) with darknet framework which is comparatively faster and accurate, contrast to existing object detectors. It gives us speedy and precise detection of text from common scene pictures. Experiments performed on state of art datasets including ICDAR 2015, ICDAR 2013, ICDAR 2003, SVT, IIIT5K, MSRA-TD500 display that the proposed pipeline significantly performs better except for ICDAR2015, compare to existing techniques with respect to correctness.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.