A Text Recognition Algorithm Based on a Dual-Attention Mechanism in Complex Driving Environment

In response to many problems such as complex background of text recognition environment, perspective distortion, shallow handwriting, and mixed Chinese and English characters, we have designed an OCR algorithm framework with features such as landmark extraction and correction, image enhancement, text detection, and text recognition. We have designed a DBNet based on dual attention mechanism and content-aware upsampling. We have also designed a text recognition module incorporating the central loss CRNN + CTC to improve content awareness. Experimental results show that the improved text detection network in this paper has increased accuracy by 5.09%, recall by 2.12%, and F-score by 3.46% on the ICDAR2015 dataset. The text recognition network has improved the accuracy of recognizing Chinese and English characters by 1.2%.