Research On Weighted Multi-manifold Ranking For The Content-based Image Retrieval

Since the 20 th century,with the development of computer network,multimedia technology,camera equipment and storage hardware,image became one of the main information carriers.In the big data era,effectively searching the needed information from the huge amounts of image data is a popular problem in the research field of artificial intelligence and machine learning.In the late of 1990 s,the Content-based Image Retrieval(CBIR)technology gradually appeared,which can analyzethe Content attribute and semantic via the related algorithm,objectively and comprehensively describe and retrievalthe query,successfully applied to the field of web Image search,medical Image analysis,security monitoring,adverse information filtering,electronic commerce and so on.Now,the content-based image retrieval technology has attracted universal attention of numerous researchers and scholars,so that many classic retrieval methods and successful retrieval systems have been put forward.Asthere are complexity and structural characteristics in retrieval problem,image feature extraction and representation and similarity measure are the key problem of these researches.In order to achieve higher retrieval performance,this paper proposes the weighted multi-manifold ranking algorithm and applied it to the content-based retrieval task,the main work as follows:1.Extract the image features based on multiple views.Image content within the class differences,especially the natural images which hasbig environmental change and noise interference phenomenon.This paper hasadopted color histogram,edge histogram and local binary pattern to describe the color,shape and texture information,to characterize the rich variety of image content from multiple perspectives.2.Puts forward multi-view efficient graph construction,which has explored the structure information of data and ensured the manifold ranking more effectively.3.Explore the effect of various weighted strategy for manifold fusion.Utilize the weighted maximum method,weighted average method and linear weighted sum methodto fuse the single-view manifold,which can mine the correlation between different perspectives.Unlike the classic retrieval methods,based on the weighted multi-manifold ranking algorithm contribute to both feature extraction and representation and similarity measure.In the benchmark of Oliva datasets,Caltech5138 datasetsand Corel1000 datasets,the experimental results verify the effectiveness of the image retrieval method of weighted multi-manifold ranking,especially fusion based on linear weighted sum method which can use the complementary information of multi-view to achieve the purpose of efficient retrieval.