干旱区生态环境知识资源中心

ALISA (Adaptive Learning Image and Signal Analysis) is an adaptive statistical learning engine that may be used to detect and classify the surfaces and boundaries of objects in images. The engine has been designed, implemented, and tested at both the George Washington University and the Research Institute for Applied Knowledge Processing (FAW) in Ulm, Germany over the last pine years with major funding from Robert Bosch GmbH and Lockheed-Martin Corporation. The design of ALISA was inspired by the multi-path cortical-column architecture and adaptive functions of the mammalian visual cortex.

Based on training with a small number of exemplars, the ALISA Texture Module has demonstrated excellent detection of anomalies and classification of textures in test images. Applications include the detection of intruders in complex environments, the detection of vehicles in desert terrain, the identification of defective motors from their acoustical signatures, the detection of aneurysms in MRI heart images, and the classification of benign and malignant 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 can be taught to recognize and classify a set of simple "canonical" geometric concepts, such as horizontal, vertical, slanted, curved, intersecting, interrupted, symmetrical, and the like. Very few training samples are required to generalize these symbolic concepts. Alternatively, the Geometry Module can learn to classify "secular" geometric concepts, such as signatures of different persons, text and graphics structures in manuscripts, different kinds of boats, open and closed doors, different kinds of coins, different kinds of fruits, different brands of batteries, different postage stamps,etc.

The ALISA Shape Module, currently under development, classifies shapes in images based on the information in geometry class maps. The Shape Module can classify canonical shape concepts, such as isoceles triangles, trapezoids, rectangles, etc., and secular shape concepts, such as faces, human bodes, Latin letters, bottles, cars, airplanes, etc.

Long-term research focuses upon the extension of ALISA to successively higher levels of cognition: textures --> geometries --> shapes --> parts --> objects --> scenes, using feedback from higher to lower levels for disambiguation, new class discovery, and the assignment of symbolic value. Toward this end, the ALISA Concept Module, which is currently under development, combines the decisions of the Texture, Geometry, and Shape Modules to classify higher level concepts. Until the Concept Module is available as a standard component of the ALISA system, the texture, geometry, and shape class maps generated by ALISA can provide useful symbolic operands for classical model-based and rule-based systems to classify more complex concepts at higher cognitive levels.