LDA单词图像表示的蒙古文古籍图像关键词检索方法

目的]为了克服传统视觉词袋方法（Bag-of-Visual-Words）中忽略视觉单词间的空间关系和语义信息等问题。方法]本文提出一种与视觉语言模型相结合的基于LDA主题模型，并采用查询似然模型实现检索。结果]实验数据表明，本文所提出的基于LDA的表示方法可以高效、准确地解决蒙古文古籍的关键词检索问题。结论]同时，该方法的性能比BoVW方法有显著提高。

LDA-Based Word Image Representation for Keyword Spotting on Historical Mongolian Documents

Objective] In order to overcome the problem of ignoring the spatial relations and semantic information between visual words in traditional visual word bag (Bag-of-Visual-Words).Methods] In this paper, a LDA-based topic model was adopted which was linearly combined with a visual language model for each word image. And the basic query likelihood model was used for realizing the procedure of retrieval.Results] Theexperimental results on our dataset showed that the proposed LDA-based representation approach could effi ciently and accurately attain to the aim of keyword spotting on a collection of historical Mongolian documents.Conclusions] Meanwhile, the proposed approach improved the performance signifi cantly than the original BoVW approach.