Constrained Tree Search for Feature Correspondence

Abstract

在電腦視覺領域中，常會遭遇到兩組特徵元件對應的問題。而利用條件限制來作對應搜尋是一種處理這類問題相當有效的技術。本篇文章首先介紹一種索引架構，可以簡化上述搜索演算法的施行。這種索引架構，主要是起因於杜威的十進位符號觀念。同時由於我們所介紹的索引架構，搜尋演算法的行為方式會如同數位式計數器，此種計數器可以很容易用硬體製成。本篇文章也對寫成的演算程式作複雜度分析。由分析的結果，我們提出一種程序，可以幫助我們未選取物體辦認時，較有效力的特徵元件及其相關的條件或限制。我們的實驗結果顯示上述搜尋演算法以及選取程序均具有實際應用的價值。

Constrained search is a useful technique for dealing with feature correspondence problems frequently encountered in machine vision. In this paper,an indexing scheme, inspired by the Dewey decimal notation, is proposed forsimplifying the implementation of the search algorithm. This scheme alsoenables the algorithm to be easily realized in hardware in view that the algorithm characterized by the scheme will behave like a digital counter. Throughanalyzing the complexity of the search algorithm, a process, that can be usedto select a robust subset of features and constraints for the purpose of objectrecognition, is presented. Experiments demonstrate the feasibilities of boththe algorithm and the process.