learning to recognize shapes
The goal of this research is to apply inductive learning methods to problems of 2D shape recognition under highly variable perceptual conditions. The multilevel logical template (MLT) methodology is being used to detect blasting caps in x-ray images of luggage. An intelligent system capable of quickly and reliably performing this task could be used to assist airport security personnel in baggage screening.
We have acquired x-ray images of luggage containing blasting caps that appear at differing degrees of occlusion and at various orientations with respect to the x-ray source. Task-oriented image transformations are used to segment blasting caps and other objects, and to extract training events, which are vectors of attribute values. These training events serve as input to the learning process which induces descriptions of shape that are robust with respect to planar rotation and translation and partial occlusion. Induced shape descriptions can be used to recognize unknown objects.
Various symbolic, non-symbolic and statistical learning approaches are being investigated for acquiring descriptions of shape, including AQ15c, neural networks, and k-nn. These learning approaches are compared using predictive accuracy, and learning and recognition time. Experimental results have demonstrated strong advantages of AQ15c over neural networks and k-nn. AQ15c also has the advantage of producing comprehensible symbolic descriptions that can be optimized by either a human or by a machine process in post-learning phases.
Maloof, M.A., Michalski, R.S.,"Learning Symbolic Descriptions of Shape for Object Recognition In X-Ray Images,"Expert Systems with Applications, 12(1), 11-20, 1997.
Maloof, M.A. and Michalski, R.S., "Learning Symbolic Descriptions of 2D Shapes for Object Recognition in X-ray Images," Proceedings of the 8th International Symposium on Artificial Intelligence, Monterrey, Mexico, October 17-20, 1995.
Maloof, M. and Michalski, R.S., "Learning Descriptions of 2D Shapes for Object Recognition and X-Ray Images," Reports of the Machine Learning and Inference Laboratory, MLI 94-4, Machine Learning and Inference Laboratory, George Mason University, Fairfax, VA, October 1994.
For more references, see publications section.