David Chen , Stanford University, Stanford
Bernd Girod , Stanford University, Stanford
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/MMUL.2013.46
Mobile visual search (MVS) systems compare query images captured by the mobile device&#x2019;s camera against a database of labeled images to recognize and augment objects seen in the device&#x2019;s viewfinder. Practical MVS systems require very fast responses to provide seamless and continuous augmentation. Thus, the recognition pipeline must be extremely efficient and reliable. Congestion on a server or slow transmissions of the query data over a wireless network would severely degrade the user experience. In this paper, we report how a memory-efficient database stored entirely on a mobile device can enable on-device queries that achieve a fast response. The image signatures stored in the database must be extremely compact to fit in the small amount of memory available on the device, capable of fast comparisons across a large database, and highly discriminative to provide robust recognition for challenging queries. First, we review two methods that use compression techniques with bag-of-visual-words histograms to generate compact and discriminative signatures. They require storage of a codebook in memory and decoding of compressed signatures during a query. In contrast, we then describe two methods based on bag-of-visual-words residuals that achieve the same retrieval performance, while using a much smaller codebook and compressed domain matching.
David Chen, Bernd Girod, "Memory-efficient Image Databases for Mobile Visual Search", IEEE MultiMedia, , no. 1, pp. 1, PrePrints PrePrints, doi:10.1109/MMUL.2013.46