5th International Mobile Multimedia Communications Conference

Research Article

Low-Rate Image Retrieval with Tree Histogram Coding

Download334 downloads
  • @INPROCEEDINGS{10.4108/ICST.MOBIMEDIA2009.7496,
        author={Vijay Chandrasekha and David M. Chen and Zhi Li and Gabriel Takacs and Sam S. Tsai and Radek Grzeszczuk and Bernd Girod},
        title={Low-Rate Image Retrieval with Tree Histogram Coding},
        proceedings={5th International Mobile Multimedia Communications Conference},
        publisher={ICST},
        proceedings_a={MOBIMEDIA},
        year={2010},
        month={5},
        keywords={},
        doi={10.4108/ICST.MOBIMEDIA2009.7496}
    }
    
  • Vijay Chandrasekha
    David M. Chen
    Zhi Li
    Gabriel Takacs
    Sam S. Tsai
    Radek Grzeszczuk
    Bernd Girod
    Year: 2010
    Low-Rate Image Retrieval with Tree Histogram Coding
    MOBIMEDIA
    ICST
    DOI: 10.4108/ICST.MOBIMEDIA2009.7496
Vijay Chandrasekha1,*, David M. Chen1,*, Zhi Li1,*, Gabriel Takacs1,*, Sam S. Tsai1,*, Radek Grzeszczuk2,*, Bernd Girod1,*
  • 1: Information Systems, Lab Stanford University.
  • 2: Nokia Research Center, Palo Alto, CA.
*Contact email: vijayc@stanford.edu, dmchen@stanford.edu, leeoz@stanford.edu, gtakacs@stanford.edu, sstsai@stanford.edu, radek.grzeszczuk@nokia.com, bgirod@stanford.edu

Abstract

To perform image retrieval using a mobile device equipped with a camera, the mobile captures an image, transmits data wirelessly to a server, and the server replies with the associated database image information. Query data compression is crucial for low-latency retrieval over a wireless network. For fast retrieval from large databases, Scalable Vocabulary Trees (SVT) are commonly employed. In this work, we propose using distributed image matching where corresponding Tree-Structured Vector Quantizers (TSVQ) are stored on both the mobile device and the server. By quantizing feature descriptors using an optimally pruned TSVQ on the mobile device and transmitting just a tree histogram, we achieve very low bitrates without sacrificing recognition accuracy. We carry out tree pruning optimally using the BFOS algorithm and design criteria for trading off classification-error-rate and bitrate effectively. For the well known ZuBuD database, we achieve 96% accuracy with only »1000 bits per image. By extending accurate image recognition to such extremely low bitrates, we can open the door to new applications on mobile networked devices.