Distinctive Features Should Be Learned
Most existing machine vision systems perform recognition based on a fixed set of hand-crafted features, geometric models, or eigen-subspace decomposition. Drawing from psychology, neuroscience and intuition, we show that certain aspects of human performance in visual discrimination cannot be explained by any of these techniques. We argue that many practical recognition tasks for artificial vision systems operating under uncontrolled conditions critically depend on incremental learning. Loosely motivated by visuocortical processing, we present feature representations and learning methods that perform biologically plausible functions. The paper concludes with experimental results generated by our method.
Unable to display preview. Download preview PDF.
- 5.E. J. Gibson and E. S. Spelke. The development of perception. In J. H. Flavell and E. M. Markman, editors, Handbook of Child Psychology Vol. III: Cognitive Development, chapter 1, pages 2–76. Wiley, 4th edition, 1983.Google Scholar
- 6.D. Marr. Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. Freeman, San Francisco, 1982.Google Scholar
- 7.S. A. Nene, S. K. Nayar, and H. Murase. Columbia object image library (COIL-20). Technical Report CUCS-005-96, Columbia University, New York, NY, Feb. 1996.Google Scholar
- 8.J. H. Piater and R. A. Grupen. Toward learning visual discrimination strategies. In Proc. Computer Vision and Pattern Recognition (CVPR’ 99), volume 1, pages 410–415, Ft. Collins, CO, June 1999. IEEE Computer Society.Google Scholar
- 15.J.-P. Thibaut. The development of features in children and adults: The case of visual stimuli. In Proc. 17th Annual Meeting of the Cognitive Science Society, pages 194–199. Lawrence Erlbaum, 1995.Google Scholar