如何在LIRE搜索中使用多特征

这是官网上的解答,先mark下来。

http://www.semanticmetadata.net/wiki/lirefaq/

==========================================================

This depends very much on the way how you combine the features. You should not add or subtract relevance scores -- not even weighted! That's because of the fact that the metric spaces, spanned by each feature, are not compatible (no real scientific explanation, but still you might get the idea). Possible approaches are

  • Result list merging, where search for each feature returns its own result list and you mash them in some way (e.g. medium rank)
  • Combined filtering and ranking, where you use one feature to select 20 results and another to rank them for a ranked result list
  • Machine learning, where you use learning methods to fit parameters and select appropriate dimensions for retrieval in your image domain
原文地址:https://www.cnblogs.com/hybingo/p/4247990.html