Accurate recognition of words in scenes without character segmentation using recurrent neural network

作者：

Highlights：

• We designed a novel method that converts a word image into a sequential signal.

• We designed an ensembling RNN for word-level scene text recognition which obtained superior recognition accuracy.

• Our method uses publicly available datasets in training which provides a baseline for benchmarking of the future works.

• Our method uses word instead character level annotations, which reduces the efforts in ground truth generation greatly.

摘要

Highlights•We designed a novel method that converts a word image into a sequential signal.•We designed an ensembling RNN for word-level scene text recognition which obtained superior recognition accuracy.•Our method uses publicly available datasets in training which provides a baseline for benchmarking of the future works.•Our method uses word instead character level annotations, which reduces the efforts in ground truth generation greatly.

论文关键词：Scene text recognition,Recurrent neural network

论文评审过程：Received 6 June 2016, Revised 20 September 2016, Accepted 15 October 2016, Available online 22 October 2016, Version of Record 28 October 2016.

论文官网地址：https://doi.org/10.1016/j.patcog.2016.10.016