ECE Graduate Seminar Lecture

Science / Technology - Lecture/Discussion - WPI Only

Thursday, March 14, 2013
11:00 AM-11:50 AM

Atwater Kent Laboratories
AK 219

Quantized Embeddings of Scale Invariant Image Features for Mobile Augmented Reality

Shantanu Rane
Principal Research Scientist
Mitsubishi Electric Research Lab (MERL), Cambridge, MA

In this talk, we will propose the use of randomized embeddings of scale-invariant image features for visual inference. As a motivating application, we consider retrieval of object-specific metadata in mobile augmented reality, and show how randomized embeddings provide a significantly superior rate-accuracy tradeoff than conventional approaches.

The principal idea is to extract-scale invariant features from a query image, compute a small number of quantized random projections of these features, and send them to a database server. The server performs a nearest neighbor search in the space of the random projections and returns meta-data corresponding to the query image. Prior work has shown that binary embeddings of image features enable efficient image retrieval. This paper generalizes the prior art by characterizing the tradeoff between the number of random projections and the number of bits used to represent each projection. The theoretical results suggest a bit allocation scheme under a total bit rate constraint: It is often advisable to spend bits on a small number of finely quantized random measurements rather than on a large number of coarsely quantized random measurements. The theoretical result is corroborated via experimental study of the above tradeoff using the ZuBud public domain database. The proposed scheme achieves a retrieval accuracy up to 94% while requiring the mobile device to transmit an average of only 2.5 kB to the database server, a significant improvement over 1-bit quantization schemes reported in prior art.

[This talk covers joint work with Petros Boufounos and Mu Li at MERL.]

Shantanu Rane (Ph.D., Stanford University, 2007) is a Principal Member of the Research Staff at Mitsubishi Electric Research Laboratories in Cambridge, Massachusetts. His research interests are in the broad areas of signal processing and information theory. His recent research projects are in the areas of secure biometrics, secure multiparty computation and distributed source coding. He has participated in standardization activity under the ITU-T/MPEGs H.264/AVC video compression standard, the US National Body for standardization of biometrics, and the ISO-SC37 Subcommittee on Biometrics. He currently serves as an associate editor for the IEEE Transactions on Information Forensics and Security, the IEEE Signal Processing Letters and the Signal Processing Society's eNewsletter. He is a member of the Signal Processing Society's Information Forensics and Security Technical Committee and a senior member of the IEEE.

Host: Professor Lifeng Lai

Suggested Audiences: College


Last Modified: February 26, 2013 at 9:59 AM

Powered by the Social Web - Bringing people together through Events, Places, & Common Interests