Peter G. Anderson, Ph.D.
Professor

Ph.D.
Massachusetts Institute of Technology

A225 Ross
(585)475-2979
E-Mail: pga@cs.rit.edu

Research Highlights

I have two connected research programs which I pursue with graduate students in Imaging Science, Computer Science, and Computer Engineering. These are the areas of neural networks and linear pixel shuffling.

Artificial neural networks are computing systems whose structure and organization are based on natural brains. These are massively parallel systems without central control: the individual computing elements ("neurons" or "cells") update their internal states asynchronously and send signals to neighboring cells for subsequent use; update functions are generally non-linear functions of a weighted sum of the inputs, or a stochastic analog of that.

Neural networks are often programmed by experience, trial and error, and ``survival of the fittest'' using genetic-style evolution, rather than by explicitly determining their parameters. They adapt to the statistics of their training data through either supervised or unsupervised learning algorithms. An emergent behavior of their massive asynchronous structure is their ability to generalize from the training data and to be immune to minor structural or parameter degradation.

We have been developing efficient neural network training algorithms and applying the neural networks to such areas as pattern recognition (hand printed and machine printed characters), image cleaning, and principal component analysis.

Linear pixel shuffling refers to the processing of the pixels in an image in an order that covers all regions in the image's area fairly and evenly from the start. We are applying this technique to resolution computer graphics color palette determination rapid approximations to morphological filters subpattern searching neural network weight determination Monte Carlo integration loss less and lossey compression of binary images

Selected Publications

Anderson, P.G., "An algebraic mask for halftone dithering," Proceedings of the 4th Annual Conference of the Society for Imaging Science &Technology, pp 487-489, May 1994.

Anderson, P.G., "Linear pixel shuffling for image processing, an introduction," Journal of Electronic Imaging, pp. 147-154, April 1993.

Anderson, P.G., Rao, A., Gaborski, R.S., and K.S. Jaiswal "A hardware polynomial feature net for hand-printed digit recognition," Proceedings of the Third IEE International Conference on Artificial Neural Networks, 36-40, 1993.

Anderson, P.G., Gaborski, R.S., Tilley, D.G., and Asbury, C.T., "Genetic algorithm selection of features for hand-printed character identification," Artificial Neural Networks and Genetic Algorithms, Proceedings of the International Conference , Innsbruck, Austria, R.F. Albrecht, C.R. Reves, and N.C. Steele (eds.) Springer Verlag, 101-106, 1993.

Anderson, P.G., Gaborski, R.S., "The polynomial method augmented by supervised training for hand printed character recognition," Artificial Neural Networks and Genetic Algorithms, Proceedings of the International Conference, R.F. Albrecht, C.R. Reves, and N.C. Steele (eds.), Springer-Verlag, pp 417-422, Innsbruck Austria 1993.