Research And Improvement Of The Image Retrieval System Based On Multi-feature Fusion

With the popular of Social Entertainment Platforms(such as WeChat,QQ),the business booming of e-commerce(such as T-mall,JD),heterogeneous data of image,video, audio and text are showing a striking growth. According to statistics, the upload pictures and the video clips has reached 1 billion per day and 20 billion per day respectively in the "moments" of WeChat. At the same time, Taobao, the largest e-commerce system, holds more than 28.6 billion pictures in its Back-End Systems.Undoubtedly, these data shows the immense value of multimedia information retrieval,especially in Image Retrieval.Recently, The technology of image retrieval has made great progress after a long period of research and development. Generally speaking, it leaps qualitative. And the change is made from TBIR(Text Based Image Retrieval) to CBIR(Content Based Image Retrieval), from manual annotation to automatically retrieve.Recently, such issue has aroused great concern among the scholars, and the problem of how to improve the retrieval accuracy has been brought into focus. So far, many improved algorithms have been proved to be effective. But, these methods still have problems in retrieval stability. In this paper, we focus on the improvement of the retrieval accuracy and the stability, which is on the basis of the existing image retrieval algorithm. The works are as follow:(1)Select three different types of features, which can extract the feature of image in maximize, and then makes the image retrieval using these features.(2)Improve the process of image retrieval which was only using three single features,the work proceeds as follow, Choose DCA as the fusion algorithm, determine the fusion scheme, and makes the image retrieval using the fusion features.(3)Select more images as the retrieve objects, establish and define the distance for every images which was using PageRank as reference. Subsequently, we making the Image retrieval results re-ranking which based on the Graph Theory.(4)Select more features as the retrieve objects, select the appropriate features(two or more)firstly, and then introducing the Sliding Window to enhance the correlation of re-ranking results, which was based on the defining the distance for every images and making the Image retrieval results re-ranking based on the Graph theory.The retrieval results show that the work which focus on the research and improvement of the image retrieval system is well completed. The method of using different types features for image retrieval, and the combination of the Graph Theory and Image Retrieval can woks stably in practical application.