TextMountain: Accurate scene text detection via instance segmentation

作者：

Highlights：

• We propose a novel method named TextMountain which is composed of TS, TCBP and TCD.

• Our experiments verify that using TCBP’s grad can greatly improve grouping text lines while TCD can help TCBP learning better.

• The proposed method can well and efficiently handle long, multi-oriented and curved scene text.

• The proposed method achieves state-of-the-art or comparable results in both quadrangle and curved polygon labeled datasets.

摘要

•We propose a novel method named TextMountain which is composed of TS, TCBP and TCD.•Our experiments verify that using TCBP’s grad can greatly improve grouping text lines while TCD can help TCBP learning better.•The proposed method can well and efficiently handle long, multi-oriented and curved scene text.•The proposed method achieves state-of-the-art or comparable results in both quadrangle and curved polygon labeled datasets.

论文关键词：Scene text detection,Curved text,Multi-oriented text,CNN,Deep learning

论文评审过程：Received 28 March 2019, Revised 9 December 2019, Accepted 12 March 2020, Available online 11 May 2020, Version of Record 1 November 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107336