Research on image text generation based on word2vec visual vocabulary attention
Danyang Li, Yahui Zhao, Rongyi Cui, Linlin Zhao · 2021
A method of image text generation based on the combination of word2vec keyword extraction and attention mechanism is proposed. First, the co-occurring words with visual entities in the description set were extracted for each image in the dataset; Then the similarity was calculated for the extracted keywords, the similar words were filtered out to expand the keyword list, and the words in the vocabulary were retained to create new descriptions for the images. Finally, the test set images were combined with attention mechanism to generate description text. The experiments prove that the method proposed in this paper can achieve automatic annotation of images and can effectively solve the attention diffusion problem in the process of image text generation.