跳到主要导航 跳到搜索 跳到主要内容

Research on Image Caption Method Based on Mixed Image Features

  • Qilu University of Technology

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

With the continuous development of deep learning in the field of image caption, the effect of it has improved. However, the traditional method for feature extraction is based on the whole picture, ignoring the local features and the relationships of global and local features. Another problem is that the text description is broad and not targeted. Considering the above problems, we propose a new method based on mixed image features. The method uses an improved ResNet to extract global features. Local features are extracted using a deep RetinaNet. The global and the local features of the image are merged by an attention mechanism, which are mapped into embedding vectors. To obtain the mapping relationship between images and descriptions, a long short-term memory (LSTM) based on an attention mechanism is used as a language generation model. Image features and semantic features are combined to generate content description of images.

源语言英语
主期刊名Proceedings of 2019 IEEE 4th Advanced Information Technology, Electronic and Automation Control Conference, IAEAC 2019
编辑Bing Xu, Kefen Mou
出版商Institute of Electrical and Electronics Engineers Inc.
1572-1576
页数5
ISBN(电子版)9781728119076
DOI
出版状态已出版 - 12月 2019
已对外发布
活动4th IEEE Advanced Information Technology, Electronic and Automation Control Conference, IAEAC 2019 - Chengdu, 中国
期限: 20 12月 201922 12月 2019

出版系列

姓名Proceedings of 2019 IEEE 4th Advanced Information Technology, Electronic and Automation Control Conference, IAEAC 2019

会议

会议4th IEEE Advanced Information Technology, Electronic and Automation Control Conference, IAEAC 2019
国家/地区中国
Chengdu
时期20/12/1922/12/19

指纹

探究 'Research on Image Caption Method Based on Mixed Image Features' 的科研主题。它们共同构成独一无二的指纹。

引用此