A robust end-to-end neural network for scene text detection

Abstract

Scene text detection is an important task in computer vision. Many previous work require multi-step processing and are not robust against challenges, such as small-scales and blurring. In this paper, we propose a Multi-level Feature extraction Network (MLFN), which can capture more detailed information and is easy to train. We embed MLFN into the middle of an end-to-end network and achieve a better accuracy and robustness. The experiment results on ICDAR 2011 and ICDAR 2013 further demonstrate the effectiveness of our method.