6th International Conference on MOBILe Wireless MiddleWARE, Operating Systems, and Applications

Research Article

TagPix: Automatic Real-time Landscape Photo Tagging For Smartphones

Download328 downloads
  • @INPROCEEDINGS{10.4108/icst.mobilware.2013.254283,
        author={Hillol Debnath and Cristian Borcea},
        title={TagPix: Automatic Real-time Landscape Photo Tagging For Smartphones},
        proceedings={6th International Conference on MOBILe Wireless MiddleWARE, Operating Systems, and Applications},
        publisher={IEEE},
        proceedings_a={MOBILWARE},
        year={2014},
        month={7},
        keywords={smartphone app phone sensors automatic photo tagging},
        doi={10.4108/icst.mobilware.2013.254283}
    }
    
  • Hillol Debnath
    Cristian Borcea
    Year: 2014
    TagPix: Automatic Real-time Landscape Photo Tagging For Smartphones
    MOBILWARE
    IEEE
    DOI: 10.4108/icst.mobilware.2013.254283
Hillol Debnath1,*, Cristian Borcea1
  • 1: New Jersey Institute of Technology
*Contact email: hd43@njit.edu

Abstract

This paper presents TagPix, a lightweight smartphone photo tagging app that provides good tagging accuracy, works in real-time, and protects user privacy. The main novelty of TagPix consists in leveraging the phone sensors and a placetag database to tag landscape photos which include landmarks. GPS location is used to identify landmark tags in a given region. Then, TagPix computes the angular distance for the object in the camera focus using the orientation sensors. This allows the app to select a small subset of landmark tags for the photo. For further tag accuracy, we devised new usable methods to estimate the Euclidean distance between the user and the landmark in the camera focus. These methods employ simple user actions and lightweight trigonometric calculations. TagPix is implemented and tested using several Android phones and Google Places API. The app was tested in 8 cities across USA. Using only angular distance, TagPix achieves 86% tagging accuracy. Adding Euclidean distance estimation leads to 93% accuracy.