Boundary-based Real-time Text Detection on Container Code
Kuikun Liu, Cai Sun, Haoyuan Chi · 2021
Scene text detection is attracting more and more attention in machine learning region. Automatic container code recognition is very important for modern container intelligent management system. However, there is no text detection network for container code. This article proposes a real-time container code text segmentation network based on boundary, which can accurately locate the text in real time. Specifically, we simultaneously predict the text and the boundary of the text, and fuse the features of the two branches to improve the accuracy of segmentation. In the post processing stage, the final result is gotten by text segmentation map minus the boundary segmentation map. We achieve a competitive F-measure of 96.5% at 70 FPS on container code datasets.