Detecting and ordering salient regions for efficient browsing

We describe an ensemble approach to learning1 salient regions from data partitioned according to the2 distributed processing requirements of large-scale sim-3 ulations. The volume of the data is such that classi-4 fiers can train only on data local to a given partition.5 Classes will likely be missing from some, or even most,6 partitions. We combine a fast ensemble learning algo-7 rithm with scaled probabilistic majority voting in order8 to learn an accurate classifier from such data. We order9 predicted regions to increase the likelihood that most of10 the initial set of presented regions are salient. Results11 from a simulated casing being dropped show that re-12 gions of interest are successfully identified and ordered.13 This approach is much faster than manually browsing14 and visualizing terabyte or larger simulations to find re-15 gions of interest. 16

• Type: Conference
• Year: 2008
• Where: ICPR
• Authors: Larry Shoemaker, Robert E. Banfield, Larry O. Hall, Kevin W. Bowyer, W. Philip Kegelmeyer
• Keywords: Computer Vision, ICPR 2008, Presented Regions, Salient Regions, Scaled Probabilistic Majority