Script identification in natural scene image and video frames using an attention based Convolutional-LSTM network

PATTERN RECOGNITION

卷:85

Article

Bhunia, Ankan Kumar¹  Konwer, Aishik²  Bhunia, Ayan Kumar²  Bhowmick, Abir²  Roy, Partha P.³  Pal, Umapada⁴ 

[1] Jadavpur Univ, Dept EE, Kolkata, India

[2] Inst Engn & Management, Dept ECE, Kolkata, India

[3] Indian Inst Technol, Dept CSE, Roorkee, Uttar Pradesh, India

[4] Indian Stat Inst, CVPR Unit, Kolkata, India

关键词: Script identification;    Convolutional neural network;    Long short-term memory;    Local feature;    Global feature;    Attention network;    Dynamic weighting;   

DOI  :  10.1016/j.patcog.2018.07.034

来源: Elsevier

【 摘 要 】

Script identification plays a significant role in analysing documents and videos. In this paper, we focus on the problem of script identification in scene text images and video scripts. Because of low image quality, complex background and similar layout of characters shared by some scripts like Greek, Latin, etc., text recognition in those cases become challenging. In this paper, we propose a novel method that involves extraction of local and global features using CNN-LSTM framework and weighting them dynamically for script identification. First, we convert the images into patches and feed them into a CNN-LSTM framework. Attention-based patch weights are calculated applying softmax layer after LSTM. Next, we do patch-wise multiplication of these weights with corresponding CNN to yield local features. Global features are also extracted from last cell state of LSTM. We employ a fusion technique which dynamically weights the local and global features for an individual patch. Experiments have been done in four public script identification datasets: SIW-13, CVSI2015, ICDAR-17 and MLe2e. The proposed framework achieves superior results in comparison to conventional methods. (C) 2018 Elsevier Ltd. All rights reserved.

【 授权许可】

Free