Design And Implementation Of Content-Based Image Retrieval System

With the deepening and expansion of application field of digital information,large-scale image sets are constantly emerging. It is urged to establish effective image representation methods and retrieval mechanisms to manage the large image database.In this context, content-based image retrieval arises and is playing a more and more important role in image resources management.The research of this dissertation focuses on the key technology of content-based image retrieval, including description methods of image content, database index,similarity matching and relevance feedback mechanism.The details are as follows:(1) The color, texture and shape features are extracted from the Corel-lk standard image library to build the feature database. The extracted features mainly include color histogram, gray-gradient co-occurrence matrix texture, LBP texture,Gabor wavelet texture and edge projection histogram.(2) Two retrieval methods are designed for users. One is the nearest neighbor query based on the establishment of Kd-tree index structure. The other is a linear search. The eigenvector distances between the example image and database images are calculated according to the similarity measure, and the images with similar characteristics to the example image will be output according to the similarity degree.(3) A novel relevance feedback approach based on glowworm swarm algorithm and SVM is proposed to bridge the semantic gap in image retrieval. It uses the intelligent optimization algorithm to automatically optimize and modify the query strategy in order to capture feedback information and improve retrieval performance.(4) An experimental content-based image retrieval system is designed and implemented on the base of Visual Studio 2013 and SQL Server 2012. This system consists of five modules: feature extraction, database, index matching, query display and relevance feedback.(5) Standard image library is used to test the functions of the CBIR system.Comparative experiments based on different features are carried out to validate effective image description methods. The retrieval performance of the system is evaluated by average precision and average response time.