Research on Automatic Image Annotation Based on Dual-branch Attention Mechanism
Automatic image annotation technology can transform low-level visual features of images into high-level semantic information understood by humans,enhancing the comprehensibility and searchability of images,and has important application value in the fields of image retrieval and classification.At present,automatic image annotation technology based on convolutional neural network models still faces problems such as shallow networks being unable to capture sufficient feature information,easily ignoring the interrelationships between labels,and difficulty in determining the number of labels during annotation.The proposed automatic image annotation method based on dual-branch attention mechanism first uses a dual-branch attention network to enhance the correlation between image features and labels,as well as learn the correlation between labels.Secondly,a multi scale feature extraction module is added to the spatial attention branch to extract multi scale features of the image,solving the problem of insufficient feature extraction in shallow networks.By fusing the outputs of the two branches again through the fusion module,the image features are further enhanced.Finally,the label quantity prediction module is used to predict the number of labels in the image to be annotated,further improving the accuracy of annotation.The proposed model was experimentally analyzed on three benchmark datasets,Corel 5K,ESP Game,and IAPR-TC-12.The experimental results showed that the proposed method can effectively solve the above problems and improve the effectiveness and ac-curacy of labeling.
国家科技期刊平台
NSTL
万方数据
