Information Retrieval Technology And System Based On Image Caption

In recent years,Image Caption is a hotspot in the field of Natural Language Processing(NLP).The main task is to generate a sentence for a picture that can be understood by people.At present,the Encoder-Decoder Framework is mostly used in this task.With proper improvement,the original Encoder-Decoder Framework based on machine translation has reached a good result.In order to make the Image Caption get better results,and let the images queried by description to be displayed on the webpage,we mainly study on the following contents:1)Implementing Image Caption with Visual AttentionWith the development of Deep Learning,the task of Image Caption is constantly being updated and improved in recent years.In this paper,we study on the model which proposed by Xu K et al.We will explain the structure and principle of the model,and proved the accuracy through experiments.2)Building image retrieval system based on LuceneA text file which contains hundreds of thousands of image descriptions may be generated after Image Caption.In order to accurately get the images that we want from this large data file,we can use the full-text search toolkit--Lucene.This paper will study on Lucene,explain its structure and principle,and build a image retrieval system based on it.