Parametric Optimization of Families of Features in Image Reconstruction Methods Based on the Principles of Pattern Recognition Theory

摘要

Designing image reconstruction algorithms based on the principles of pattern recognition theory (PR-methods) involves two stages: forming a system of local features on an image and finding optimal coefficients of approximation of an ideal image by this system of features. All initial features or a part of them may continuously depend on the parameters (such as the radius of the processing window). In this case, an iterative procedure for joint optimization of the parameters of the selected features and the approximation coefficients of the ideal image is constructed. The developed algorithm can also be used to successively generate a parametric family of features.