Title:
METHOD FOR GENERATING IMAGE SEMANTIC DESCRIPTION
Document Type and Number:
WIPO Patent Application WO/2020/244287
Kind Code:
A1
Abstract:
Provided is a method for generating an image semantic description, belonging to the technical field of image semantic description. The method comprises the following steps: extracting a global feature and a local feature of an image (S1); combining the global feature and the local feature to obtain combined information (S2); by means of an adaptive attention mechanism and according to the combined information, focusing the image features and focusing text to generate a semantic description model; using the semantic description model to generate image semantic description words and sentences (S3). The method solves the problem, in current image semantic description technology, of determining the focus point of an image, mining higher-level semantic information, and improving the detail information of descriptive words or sentences.
Inventors:
ZHAO XIAOHU (CN)
YOU PENG (CN)
YIN LIANGFEI (CN)
ZHANG ZHIQIANG (CN)
SHEN XIUQI (CN)
LIU YONG (CN)
YOU PENG (CN)
YIN LIANGFEI (CN)
ZHANG ZHIQIANG (CN)
SHEN XIUQI (CN)
LIU YONG (CN)
Application Number:
PCT/CN2020/081287
Publication Date:
December 10, 2020
Filing Date:
March 26, 2020
Export Citation:
Assignee:
UNIV CHINA MINING (CN)
International Classes:
G06N3/04; G06K9/46
Foreign References:
| CN110188779A | 2019-08-30 | |||
| CN110210499A | 2019-09-06 | |||
| CN108268875A | 2018-07-10 | |||
| CN108985370A | 2018-12-11 | |||
| CN101706780A | 2010-05-12 |
Attorney, Agent or Firm:
NANJING JINGWEI PATENT & TRADEMARK AGENCY CO., LTD (CN)
Download PDF: