Deirdre B. O'Brien

Learn More
Recently developed optical inspection tools provide images from the inside of natural gas pipelines to monitor pipeline integrity. The vast amounts of data generated prohibits human inspection of the resulting images. We designed an image processing and classification method to identify abnormal events. Non-overlapping image blocks are classified into(More)
For two-class classification, it is common to classify by setting a threshold on class probability estimates, where the threshold is determined by ROC curve analysis. An analog for multi-class classification is learning a new class partitioning of the multiclass probability simplex to minimize empirical misclassification costs. We analyze the interplay(More)
Gauss mixture (GM) models are frequently used for their ability to well approximate many densities and for their tractability to analysis. We propose new classification methods built on GM clustering algorithms more often studied and used for vector quantization (VQ). One of our methods is an extension of the 'codebook matching' idea to the specific case of(More)
Previous work has shown Gaussian mixture vector quantization (GMVQ) based classifiers to be effective in classifying image blocks, image regions and whole images. A significant attraction of GMVQ for whole image classification is that simple local features can be used, thereby avoiding time-consuming feature design and selection. Unfortunately, however,(More)
The Wasserstein distortion has proved useful in a variety of mathematical, signal processing and coding problems as a measure of how different two distributions are. In this paper we provide an expression for the performance of the optimal entropy constrained quantizer in terms of the Wasserstein distortion. The proof presented is significantly different(More)
  • 1