Summary-Based Efficient Content Based Image Retrieval in P2P Network

The World Wide Web provides an enormous amount of images which is generally searched using text based methods. Searching for images using image content is necessary to overcome the limitations of text based search. Generally, in Unstructured P2P systems like Gnutella a complete blind search is used that floods the network with high query traffic. In this paper, we present a P2P system that uses "`informed search"' in which peers try to learn about the information maintained at their neighbours in order to minimise the query traffic. Here the images are first clustered using K-means clustering technique and then each peer is made to exchange its cluster information with its neighbouring peers using PROBE and ECHO. Typically one summary table per peer is maintained in which neighbouring peers data information is stored. When processing queries, these summaries are used to choose the most probable peer that is likely to contain information relevant to the query. If none of its neighbours has a match then standard random-walk algorithm is used for query propagation.