关键词: 视觉词袋模型, 模型选择, 高斯混合模型, 贝叶斯信息准则

Abstract: Visual Bag-of-Words model has been widely used in image classification，retrieval and recognition.However，its word number usually is selected by user experience or determined using the supervised cross-validation scheme.In this paper，an unsupervised method is proposed to infer the word number of Visual Bag-of-Words model（BoW） based on the idea of model selection.Firstly，Gaussian Mixture Models（GMM） are built accounting for BoWs with different word number.Afterwards，Bayesian Information Criterion（BIC） is adopted to select the best model that has the minimum BIC value.Compared with cross-validation approach using image classification，the result demonstrates the effectiveness of the proposed approach.

Key words: Visual Bag-of-Words, model selection, Gaussian Mixture Model（GMM）, Bayesian information criterion