A New Strategy for Reducing Errors in Scene Text Detection

Compared to regression based text detection methods, segmentation based methods can better handle irregularly shaped text regions and effectively reduce detection errors. They can flexibly cope with different text forms and better handle the problem of multilingual text detection. However, due to the limitations of segmentation algorithms themselves, segmentation based methods may result in significant errors in situations where the boundaries between text and non text regions are blurred. In this article, we propose a novel strategy to address a series of inaccuracies and errors that arise during the detection process. This method also simplifies some tedious operations during the detection process, thereby improving the overall performance of the model. We have conducted multiple experiments on multiple common benchmarks to verify the performance and effectiveness of our proposed model, which has reached the level of state-of-the-art methods in terms of accuracy and speed in detection results.