Not Logged In

Learning robust object recognition strategies

Automated image interpretation is an important task innumerous applications ranging from security systems tonatural resource inventorization based on remote-sensing.Recently, a second generation of adaptive machinelearnedimage interpretation systems have shown expertlevelperformance in several challenging domains. Whiledemonstrating an unprecedented improvement over handengineeredand first generation machine-learned systems interms of cross-domain portability, and design-cycle time,such systems have yet to be rigorously tested. This paperinspects the anatomy of the state-of-the-art Multi ResolutionAdaptive Object Recognition framework (MR ADORE)and presents experimental results aimed at establishing therobustness of the system to real-world image perturbations.Tested in a challenging domain of forestry, MR ADORE isshown to be robust to changes in sun angle, camera angleand training signal accuracy.

Citation

I. Levner, V. Bulitko, L. Li, G. Lee, R. Greiner. "Learning robust object recognition strategies". Eighth Australian and New Zealand Conference on Intelligent Information Systems (ANZCIIS), New Zealand, pp 489-494, December 2003.

Keywords: Adaptive and Machine Learning, Intelligent Image Processing and Computer Vision, MrAdor, machine learning
Category: In Conference

BibTeX

@incollection{Levner+al:ANZCIIS03,
  author = {Ilya Levner and Vadim Bulitko and Lihong Li and Greg Lee and Russ
    Greiner},
  title = {Learning robust object recognition strategies},
  Pages = {489-494},
  booktitle = {Eighth Australian and New Zealand Conference on Intelligent
    Information Systems (ANZCIIS)},
  year = 2003,
}

Last Updated: May 29, 2007
Submitted by Staurt H. Johnson

University of Alberta Logo AICML Logo