The Research Of Image Retrieval Method Constructed By Clustering Index

With the rapid development of multimedia technology,the size of the image library becomes larger and larger,so that the image retrieval technology has become an important field of computer vision branch.Image Retrieval related technology research has experienced three stages: text-based,content-based and semantic-based retrieval,current research is mainly focus on semantic-based image retrieval.Based on the using of content-based retrieval,semantic-based retrieval integration of machine learning,artificial intelligence,pattern recognition,data mining and other knowledge,to enhance the ability to image retrieval.However,the current image retrieval accuracy remains to be further enhanced,so this thesis will be integrated cluster analysis and relevance feedback to reduce the difference between users understanding of the actual image and content-based image retrieval search results that failed to take into account of the actual semantic information and to build the index structure for the image library to eventually reduce query time,improve the accuracy of image retrieval.This thesis research on image retrieval method for clustering index build,the main works are as follows:Firstly,the existing low-level visual features of images are often applied on only one aspect of the graphs that have been described and the features is only a reflection of image surface in a certain laws without actual semantic information,which can cause a large amount of image information loss.This thesis made some improvements on the traditional Hu moment invariants and local color feature,increase the ability of the expression of characteristics,add a number of semantic information to make it more precision.Secondly,building a hierarchical index based on cluster analysis,with cluster analysis to tissue the shape and visual characteristics that extracted from image library.Two features formed index structure.Shape features are used for a larger particle size query and color feature is used in further detail retrieval to improve the accuracy and diversity characteristics of the query.Thirdly,Study on the classification that based on user feedback process.Focus on users' feedback of image retrieval,machine learning through the use of SVM algorithm result of user feedback between image features and advanced underlying semantics of training and learning,human-computer interaction and reduced completion the distance so that the image search results closer to the point of interest of the user,more accurate.