Recovering Images Based On Their Contents In A Distributed System

Abstract

The use of peer-to-peer networking as an alternative has become more common in recent years as a means of facilitating the scalable movement of multimedia data. The process of carrying out content-based retrieval in peer-to-peer networks, which are characterized by the distribution of huge quantities of visual data across several nodes, is an important yet challenging topic. In this study, we offer a scalable strategy for content-based picture retrieval in peer-to-peer networks utilizing the bag-of-visual-words paradigm. This is in contrast to most of the previous approaches, which were focused on indexing high-dimensional visual characteristics and had limitations on their scalability. When images are scattered over the whole of the peer-to-peer network, the key challenge lies in efficiently getting a global codebook. This is not a problem in centralized setups because it is easier to access the codebook. A static codebook is less helpful for retrieval tasks in a peer-to-peer network because of the dynamic nature of the growth of the network itself. In order to accomplish this, we present a method for dynamically updating the codebook. This method works by distributing the workload evenly across the nodes that are responsible for handling different code words and optimizing the mutual information that exists between the generated codebook and the relevance information. In order to speed up the retrieval process and cut down on network overhead, researchers are investigating several methods for index trimming. The comprehensive experimental data that we have collected indicates that the method that has been recommended is scalable in dynamic and scattered peer-to-peer networks, all while improving retrieval accuracy.