计算机应用与软件2018，Vol.35Issue(1)：211-216,231,7.DOI:10.3969/j.issn.1000-386x.2018.01.037
图像语义相似性网络的文本描述方法
IMAGE CAPTION BASED ON IMAGE SEMANTIC SIMILARITY NETWORK
摘要
Abstract
Image caption solves the problem of high level semantic image understanding.Due to the semantic gap,there are differences between the generative captions and images,and the shallow neural network based language model is hard to generate smooth sentences.Tosolve this problem,this paper proposes image semantic similarity neural network,which adds full connected layers after the output layer of recurrent neural network,and thus introduce the visual similarity and semantic similarity information between images.Besides,this method increases the depth of stacked hidden layers and common hidden layers to improve the learning ability of language model,and finally obtain image captions close to human natural language.Experiments show that the proposed method get great respond from BLEU,ROUGE,METEOR and CIDEr assessment criteria by obtaining high quality captions.关键词
图像文本化描述/递归神经网络/语义相似性/语言模型/语义鸿沟/束搜索Key words
Image caption/Recursive neural network/Semantic similarity/Language model/Semantic gap/Beam search分类
信息技术与安全科学引用本文复制引用
刘畅,周向东,施伯乐..图像语义相似性网络的文本描述方法[J].计算机应用与软件,2018,35(1):211-216,231,7.基金项目
国家自然科学基金项目(61370157). （61370157）
