Research On Image Retrieval Based On Image Caption And Automatic Annotation

Most of the current image retrieval are text-based image retrieval or content-based image retrieval.The former has the problem of manual labeling,and the latter has the problem of semantic gap.This paper uses deep learning to automatically label images,and converts the extracted high-level semantic information into rich labeling information,which not only eliminates the problem of semantic gap,but also saves a lot of manpower costs and enables efficient image retrieval.According to the different implementation methods of automatic annotation,the research content of this paper can be divided into two parts: first,image retrieval based on image classification model and automatic annotation,that is,a multi-label classification model is used to automatically annotate one or more label information to the picture,users can search image by keywords of category.Then analyzed the advantages and disadvantages of using the image classification model for automatic annotation.Based on this,this paper proposed image retrieval based on the image caption model and automatic labeling.That is,the image caption model was used to automatically annotate descriptive text to the picture.Users can enter descriptive information for image retrieval.Compared with the classification model,the image caption model can generate richer annotations and can retrieve the target picture more quickly.This article uses Django-based web framework Django to build a complete image retrieval system based on image caption models and automatic annotation.It can automatically generate descriptive text for pictures uploaded by users.The text information entered by the user performs similarity calculation with the description information in the database,returns one or more pictures with the highest similarity,and completes the image retrieval.