Handwritten digit segmentation: Is it still necessary?

作者：

Highlights：

• Modular system without segmentation to recognize unconstrained numerical strings.

• It uses representation learning for string length classification and digit classification.

• It outperforms all segmentation algorithms available in the literature on the Touching Pairs Database.

• It achieves state-of-the-art performance on NIST SD19 dataset.

摘要

•Modular system without segmentation to recognize unconstrained numerical strings.•It uses representation learning for string length classification and digit classification.•It outperforms all segmentation algorithms available in the literature on the Touching Pairs Database.•It achieves state-of-the-art performance on NIST SD19 dataset.

论文关键词：

论文评审过程：Received 3 January 2017, Revised 31 December 2017, Accepted 7 January 2018, Available online 10 January 2018, Version of Record 30 January 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.01.004