Visual Search: A Large-Scale Perspective

摘要: Abstract Advances in the proliferation of media devices as well Internet technologies have generated massive image data sets and made them easier to access share today. These large-scale sources not only provide rich test beds for solving existing computer vision problems, but also pose a unique challenge processing that demands an effective information retrieval system browse search. This is motivated by many real-world applications where visual search has been shown offer compelling interfaces functionalities capturing attributes better than modalities such audio, text, etc. In this chapter, we describe state-of-the-art techniques Specifically, outline each phase typical pipeline including extraction, representation, indexing, matching, focus on practical issues memory footprint speed while dealing with large sets. We tabulate several public commonly used benchmarks, along their summary. The scope reveals wide variety potential vision-based system. Finally, address promising research directions introducing some other core components serve improve current