Face recognition using Principal Component Analysis and Potts Models
| Autor: | Yi-Hsiang Yang, 楊益祥 |
|---|---|
| Rok vydání: | 1999 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 87 In case of face recognition, there are several advances which have employed forms of Principal Component Analysis ( PCA ). In this work, we use PCA to extract features as a preprocessing stage, and then propose a novel neural network on the basis of interactive Potts models for discriminant analysis. As a sparse and distributed encoding, Potts models consisting of multi-stage neural variables and receptive fields form internal representations for nonlinear and complicate boundary structures within the input parameter space of a classification task. An optimization mathematic framework, which contains a set of objectives and constrains, characterizes Potts models for discriminant analysis and leads to a Hopfield-like energy function. Since there exist tremendous local minimum within the new energy function, by using a hybrid relaxation of the mean field or deterministic annealing and the gradient descent method, multiple sets of interactive dynamical equations are derived as the evolution of Potts neural variables and corresponding receptive fields for easily obtaining near global minimum of the energy function. New Potts models carry out parallel and distributed computations for discriminant analysis and show encourage performance in numerical simulations. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|