Research On Image Feature Labeling For Query Schemes In Storage Systems

With the rapid growth of data size and difficult processing, big data becomes moreand more important. Storage systems cannot support image files and other unstructuredquery services for big data. Traditional file metadata information does not demonstrate thecontents of image files and has low differentiation. Image files in storage systems are fastgrowing and facing the challenges of data size and query performance. Thus, an indexscheme is proposed for managing image files in distributed storage systems. It caneffectively improve image file query efficiency and provide the accurate query.The design and implementation in a distributed storage system for image file querycan support image file based queries for users’ requests. File metadata are used to supportthe analysis of speeded-up robust schemes to extract image feature labeling and enhanceintensity of image expression. Locality sensitive hashing algorithm is used to address theproblem of nearest neighbor searching based on image feature labeling. The main contentsinclude the combination of local sensitive hashing and MapReduce programming model.A distributed index construction scheme is used to leverage the image feature labeling andfurther generate index records in a distributed environment. Time and space overhead canbe reduced and the query efficiency of image files is improved. Finally, by separating theindex structure and user query, a relational database is used to store the index informationof image file and further reduce the response time of user’s query requests.The results show that the distributed index construction and image feature labelingcan significantly improve system performance, query latency and query accuracy.