Design Of An Image Retrieval System That Considers Ambiguity And Diversity

With the popularity of social networking platforms and mobile intelligent terminals,image,video data has become a real big data.This is convenient for everyone’s daily life and work.However,this is also a double-edged sword.Faced with such a mass of data,it is increasingly difficult for users to quickly access to the required information.In order to obtain the required information quickly and efficiently,image retrieval is becoming the main tool for everyone to search for information.Because of the ambiguity of semantic gap and keywords itself,image system can not accurately capture the user’s retrieval intention.On the other hand,with the explosive growth of data,the search results often have too many similar images.Therefore,the image retrieval systems need to take into account the ambiguity and diversity.Based on this,the image retrieval results as the goal,for the lack of diversity and ambiguity of content retrieval,the specific work is divided into the following sections:1)This paper proposes a algorithm for diversity re-ranking based on topic coverage.The algorithm fully combines the semantic and visual information of the image to analyze the correlation between the image and the retrieval content and the diversity between the image and the sorted list.At the same time,the algorithm uses the value of density peaks of the image to analyze the topic coverage ability of the image,and it is introduced into the re-ranking process.Finally,the proposed algorithm is used to re-rank the results so that the results are more diversified and include more topics on the basis of relevance.2)An interactive feedback framework is proposed.Firstly,the K-Means clustering algorithm is improved by using the value of density peaks of the image.Then,the diversity adjustment method is proposed,which combines the results of clustering and the re-ranking algorithm,and the user can choose the simple parameter to achieve the diversity of the results.Finally,the relevance feedback method based on the click interaction is proposed.Users can acquire the most relevant images to the clicked image using only one click.3)Designing and implement the image retrieval prototype system.The proposed system can effectively combine the proposed algorithm and feedback framework,and provides users with a good interactive interface to browse and search results.