method: Attention based OCR2019-04-19

Authors: Anh Duc Le, Dung Van Pham, Tuan Anh Nguyen

Description: We propose a text detection and recognition based on Connectionist Text Proposal Network (CTPN) and Attention-based Encoder-Decoder (AED). The proposed method effectively detects horizontal text by CTPN and recognize each text line by ADE. The pre-processing to extract receipt area was employed to improve the text detection. CTPN and ADE were trained separately. ADE provided the final result for each detected text line.

1. Anh Duc Le, Bipin Indurkhya, Masaki Nakagawa, Pattern Generation Strategies for Improving Recognition of Handwritten Mathematical Expressions, https://arxiv.org/abs/1901.06763 2. Anh Duc Le, Hung Tuan Nguyen, and Masaki Nakagawa , Recognizing Unconstrained Vietnamese Handwriting By Attention Based Encoder Decoder Model, International Conference on Advanced Computing and Applications, Nov 2018.