Volltext-Downloads (blau) und Frontdoor-Views (grau)

Automatic identification of focus personage in multi-lingual news images

  • Annotations of character IDs in news images are critical as ground truth for news retrieval and recommendation system. Universality and accuracy optimization of deep neural network models constitutes the key technology to improve the precision and computing efficiency of automatic news character identification, which is attracting increased attention globally. This paper explores the optimized deep neural network model for automatic focus personage identification in multi-lingual news. First, the face model of the focus personage is trained by using the corresponding face images from German news as positive samples. Next, the scheme of Recurrent Convolutional Neural Network (RCNN) + Bi-directional Long-Short Term Memory (Bi-LSTM) + Conditional Random Field (CRF) is utilized to label the focus name, and the RCNN-RCNN encoder–decoder is applied to translate names of people into multiple languages. Third, face features are described by combining the advantages of Local Gabor Binary Pattern Histogram Sequence (LGBPHS) and RCNN, and iterative quantization (ITQ) is used to binarize codes. Finally, a name semantic network is built for different domains. Experiments are performed on a dataset which comprises approximately 100,000 news images. The experimental results demonstrate that the proposed method achieves a significant improvement over other algorithms.

Download full text files

  • 3066.pdf

Export metadata

Additional Services

Search Google Scholar


Author of HS ReutlingenRätsch, Matthias
Erschienen in:Multimedia tools and applications : an international journal
Publisher:Springer Science + Business Media
Place of publication:Dordrecht
Document Type:Journal article
Publication year:2021
Tag:iterative quantization (ITQ); long short-term memory network (LSTM); multi-lingual news; recurrent convolutional neural network (RCNN)
Page Number:16
First Page:11015
Last Page:11030
DDC classes:070 Nachrichtenmedien, Journalismus, Verlagswesen
Open access?:Nein
Licence (German):License Logo  In Copyright - Urheberrechtlich geschützt