Image abstraction for improved semantic retrieval accuracy and reduced space-time complexities

There have been many developments in content-based image retrieval (CBIR) recently. While retrieval quality of such systems has improved dramatically in the last decade, the complexities involved made most available systems impractical in real-life use. These complexities are a combination of space use and computational time required. Most approaches available to reduce these two complexities are a trade-off with retrieval accuracy and quality. In this work, we demonstrate a novel abstraction system based on Singular Value Decomposition (SVD), which is able to reduce the complexities of image retrieval systems, while at the same time enhancing the retrieval quality. We used an implementation of a very recent content-based image retrieval system named IPSILON [11] to demonstrate how quality and complexity are greatly enhanced.