相似索引等距包络参数计算的改进算法

Envelope parameter calculation of similarity indexing structure

在相似索引等距包络（球包络）的参数计算中，直接计算方法由于计算代价过高而难于应用。 Ｒ． Ｋｕｎｉａｗ ａｔｉ和 Ｊ． Ｓ． Ｊｉｎ 针对欧氏空间情形提出一种迭代的 γ空间搜索算法，但其计算过程需要保存前面计算得到的所有平面参数，在实际应用中受到一定限制。为了解决这个问题，该文对γ空间搜索算法进行了改进，避免了原算法的缺点，并将改进算法进一步推广到二次型距离空间和街区距离空间中。文中给出了算法的基本思想，以及必要的定理证明。

In similarity indexing, direct calculation of equidistant envelope (bounding sphere) parameter is difficult because of its complexity in high dimension space. The γ spatial search algorithm proposed by R. Kuniawati and J.S.Jin solved the problem in Euclidean distance space, but it needs preserve all past plane parameters, which limit its usage. In this paper, an improved algorithm to the original γ spatial search algorithm was presented, which is more simple and more practical in real applications. Besides the Euclidean distance space, algorithms in other distance spaces (including quadric distance space and block distance space) were also discussed and presented. Basic thoughts and theorems were provided, also with implementation in real content based retrieval system on large image database. The improvement on realtime performance of query response proves the efficacy of the algorithm.