Publications

IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops

Koudelka, Melissa L.; Dorsey, Daniel J.

In real-world object identification systems, the operational mission may change from day to day. For example, a target recognition system may be searching for heavy armor one day, and surface-to-air assets the next, or a radiation detection system may be interested in detecting medical isotopes in one instance, and special nuclear material in another. To accommodate this 'mission of the day' type scenario, the underlying object database must be flexible and able to adjust to changing target sets. Traditional dimensionality reduction algorithms rely on a single basis set that is derived from the complete set of objects of interest, making missionspecific adjustment a significant task. In this work, we describe a method that uses many limited-size individual basis sets to represent objects of interest instead of a single unifying basis set. Thus, only the objects of interest for the mission at hand are used at any given time, and additional objects can be added to the system simply by training a basis for the new object. We demonstrate the modular identification system on the problem of identifying radioisotopes from their gamma ray spectra using nonnegative matrix factorization.

Koudelka, Melissa L.; Dorsey, Daniel J.

Abstract not provided.

Koudelka, Melissa L.; Dorsey, Daniel J.

Abstract not provided.

Koudelka, Melissa L.; Koch, Mark W.; Russ, Trina D.

Face recognition systems require the ability to efficiently scan an existing database of faces to locate a match for a newly acquired face. The large number of faces in real world databases makes computationally intensive algorithms impractical for scanning entire databases. We propose the use of more efficient algorithms to 'prescreen' face databases, determining a limited set of likely matches that can be processed further to identify a match. We use both radial symmetry and shape to extract five features of interest on 3D range images of faces. These facial features determine a very small subset of discriminating points which serve as input to a prescreening algorithm based on a Hausdorff fraction. We show how to compute the Haudorff fraction in linear O(n) time using a range image representation. Our feature extraction and prescreening algorithms are verified using the FRGC v1.0 3D face scan data. Results show 97% of the extracted facial features are within 10 mm or less of manually marked ground truth, and the prescreener has a rank 6 recognition rate of 100%.