力学系としてのハフ変換

書誌事項

タイトル別名
  • Dynamics for the Hough Transform

この論文をさがす

抄録

ハフ変換を最小自乗推定問題として定式化する. 次いで, この性質を利用して平面上の多数の直線を抽出する力学系を導き, ハフ変換が力学系の一つの計算法であることを示す. また, この力学系の性質を, 代数曲線の標本点からの復元や多面体の復元にも拡張できることを示す. さらに, これらの推定問題が半正定値計画問題になっていることを明らかにする.
The least-squares method (LSM) efficiently solves the model fitting problem, if we assume a model equation. However, for the fitting for a collection of models, the classification of data is required as pre-processing. We show that the random sampling and voting method, which is an extension of the randomized Hough transform, achieves both the calssification of sample points and the model fitting concurrently. Furthermore, we express the classification of data for the model fitting problems in machine vision as the permutation of matrices which are defined by data. We show that these frameworks derive the energy minimization problems. Furthermore, we derive a class of the dynamical systems based on the gradient flow. The present paper also shows the relations among the principal component analysis (PCA) as a method to solve LSM, the Kohonen's self organization map (SOM), a class of dynamical system, and the semidefinite programming problem (SDP). Our SOM permits algebraic manifold as feature space, that is, our accumulator space are manifolds based on the algebraic properties of the parameters which should be estimated.

収録刊行物

参考文献 (15)*注記

もっと見る

詳細情報 詳細情報について

  • CRID
    1571698602393600512
  • NII論文ID
    110003274908
  • NII書誌ID
    AN10541106
  • 本文言語コード
    en
  • データソース種別
    • CiNii Articles

問題の指摘

ページトップへ