Beyond standard regularization theory
A set of local interaction field are suggested to replace the δ error term in usual regularization approaches. These local fields bring some computational and conceptual benefits. A set of local oriented position pinning fields and orientation tuning fields are suggested to use local position and orientation correlations directly in regularization. Some simple experiments show that these generalizations are useful in many cases.
KeywordsPrincipal Curvature Markov Random Field Surface Patch Orientation Data Regularization Theory
Unable to display preview. Download preview PDF.
- 1.Poggio, T., Girosi, F.: Regularization algorithms for learning that are equivalent to multilayer networks. Science 24 (1990) 978–982Google Scholar
- 2.Li, S. Z.: Markov Random Field Modeling in Computer Vision. Springer-Verlag 1995Google Scholar
- 3.Poggio,T.: A theory of how the brain might work. Cold Spring Harbor Sym. on Quantitative Bio. (1990)978–982Google Scholar
- 4.Zhiyong Yang, Songde Ma: Local Interaction Fields and Adaptive Regularizers For Surface Reconstruction and Image Relaxation. Technical Report, NLPR, Inst. of Automation, Chinese Acad. of Sci.(1996)Google Scholar
- 5.Safran, S. A.: Statistical Thermodynamics of Surfaces, Interfaces, and Membranes. Addison-Wesley Pub. Com.(1994)Google Scholar
- 6.Zhiyong Yang, Songde Ma: Combine local oriented position and orientation correlations with regularization. Technical Report, NLPR, Inst. of Automation, Chinese Acad. of Sci. (1996)Google Scholar