A Method of Dynamically Recovering Feature Dimensions for Content-based Image Retrieval

摘要

Peer-to-peer processing, where the direct exchange of services or data is done between computers, is expected to be as important to the future of distributed network systems. When image retrieval initiated from one computer is executed by other computers, a large part of the computation is consumed in extracting the features of the image data. This paper proposes a method to reduce the amount of computational load necessary for execution at each computer. This reduction is accomplished by using only a part of the features, from which the whole set of features is interpolated by associative memory based on the Kohonen Map. This idea is tested with simple tasks of image retrieval.