基于高斯统计模型的场景分类及约束机制新方法  

A Novel Method for Scene Categorization with Constraint Mechanism Based on Gaussian Statistical Model

作　　者：谢昭[1] 高隽[1] 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《电子学报》2009年第4期733-738,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60375011;No.60875012);新世纪优秀人才支持计划(No.NCET-04-0560)

摘　　要：针对场景分类问题,提出了一种基于高斯统计模型的方法,无需手动标记特征表示的场景类别信息,通过无监督聚类和高斯模型的参数学习建立场景语义和"基元"信息的概率对应关系,实验结果说明,提出的方法较其他有监督的场景分类方法高效,与复杂层状模型下的分类结果非常近似.同时,初步研究了场景分类对目标分析的约束机制,通过实验证明场景信息能为目标分析提供有效的先验知识,限定不同场景下目标出现的可能性,提高了目标分类的准确性.Focusing on scene categorization, we presented a method based on Gaussian statistic model. Without feature handannotation for scene category, we can use unsupervised clustering and parameter estimation in Gaussian model to build corresponding probabilistic relations between scene labelling semantics and ＂textons＂ information. The experiments reveal that the proposed method performs more efficiently than other supervised ones and has the quite approximate results with complex hierarchical models. At the same time, we also studied constraint mechanism for object analysis primarily. Finally, we can draw conclusions that scene information provide efficient prior knowledge for object analysis by experiments. With constraints on possibilities of the object appearance in different scenes, object accuracy can be improved.

关 键 词：场景分类 高斯统计模型 参数学习 约束机制 目标分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]