After training with a few known exemplars, the ALISA Texture Module demonstrates robust detection of anomalies and classification of textures in test images and signals. Several applications of the Texture Module have been developed, including the detection of vehicles in desert terrains, the identification of defective motors from their acoustical signatures, the detection of intruders in restricted environments, the classification of fingerprints, speaker-independent classification of phonemes in continuous speech, the detection of aneurysms in MRI heart images, and the classification of healthy and pathological cells in liver samples and mammograms.
The ALISA Geometry Module classifies geometric structures in images by extracting categorical features from the texture class maps generated by the Texture Module. In one mode, the Geometry Module learns to recognize and classify a set of fundamental canonical geometric concepts, such as horizontal, vertical, slanted, curved, intersecting, interrupted, symmetrical, and the like. In another mode, the Geometry Module can be used to learn and classify secular geometric concepts, such as signatures of different persons, open and closed doors, text and graphics structures in documents, different kinds of boats, different types of coins, different kinds of fruits, and different brands of batteries. Very few training samples are required to generalize these symbolic concepts.
ALISA Texture and Geometry modules can be arranged in parallel ranks and sequential layers in a variety of configurations for specific applications, including the simultaneous processing of combinations of images and signals from different sources for sensor fusion.
Current research here at the GWU is focused on the design of the Component Module, a fundamental building block of the proposed hierarchical cognitive architecture. This development is currently funded by DTRA (Defense Threat Reduction Agency in the DoD). Also under development is the Lexical Module for language undertsanding. Long-term research is directed toward the design and implementation of this architecture, extending ALISA to successively higher levels of adaptive cognition: textures --> geometries --> shapes --> components --> objects --> scenes, using feedback from higher to lower levels for disambiguation, selective attention, and the assignment of emotional value.
The GWU research team is directed by Professor Peter Bock with a small group of doctoral students, who will finish their doctorates just before Professor Bock retires from GW in May 2008. However, the ALISA community spans many industries and government agencies here and abroad including APL, Unisys, Lockheed Martin, MITRE, LANIA (Mexico), Robert Bosch (Germany), NSA, and, of course, DTRA. The intellectual property of the ALISA method is held by Robert Bosch GmbH in Stuttgart, Germany, and The ALIAS Corporation here in DC.