A comprehensive comparison of end-to-end approaches for handwritten digit string recognition

作者：

Highlights：

• A comprehensive comparison of end-to-end approaches applied to this task.

• A string of digits is represented as a sequence of objects.

• It produces a short pipeline when compared to the traditional approaches.

• The object-detection approach compares favorably achieving state-of-art performance.

• The ground-truth annotation and the lack of context are the bottlenecks.

摘要

•A comprehensive comparison of end-to-end approaches applied to this task.•A string of digits is represented as a sequence of objects.•It produces a short pipeline when compared to the traditional approaches.•The object-detection approach compares favorably achieving state-of-art performance.•The ground-truth annotation and the lack of context are the bottlenecks.

论文关键词：Handwritten digit string recognition,Handwritten digit segmentation,Convolutional neural networks,Deep learning

论文评审过程：Received 17 December 2019, Revised 25 September 2020, Accepted 29 October 2020, Available online 9 November 2020, Version of Record 20 November 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.114196