Automatic Image Annotation Based On Deep Neural Network And Nearest Neighbor Model

In the era of big data and the Internet,massive amounts of image data are uploaded to the Internet by users at all times.Although the amount of image data is huge and disorderly,it contains a lot of valuable information.In order to effectively manage these image data and dig out the valuable information in it,automatic image annotation technology emerges from time to time.The purpose of automatic image annotation is to add appropriate semantic tags to the image through the understanding of the high-level semantics of the image.Automatic image annotation is a research hotspot in the field of image processing at present,and it has a wide range of applications in many fields and disciplines.The main work of this paper is as follows:1.An image automatic annotation model based on deep neural network and nearest neighbor model is proposed.This model divides the image automatic annotation work into two stages: image feature matrix construction and semantic annotation.2.In the stage of constructing the feature matrix of the neighbor image,using the idea of the neighbor model,an algorithm is proposed to combine the feature matrix weight of the image to be labeled with its neighbor image into the feature matrix of the neighbor image.The shallow visual features of the image extracted by the residual network pre-training model are combined with the deep visual features,so that the constructed neighbor image feature matrix covers more comprehensive image features and can better characterize the image.3.In the semantic annotation stage,a deep neural network model incorporating 3D convolution operation and self-attention mechanism function is proposed.This model can extract visual features from the labeled image and its neighbor images at the same time,and then mine the feature correlation between neighbor images,Improve the accuracy of image annotation.Experiments were carried out on Corel5 K and PASCAL VOC 2012 data sets and compared with other models.The results show that the image annotation model proposed in this paper has obtained more accurate annotation results on these two data sets.