基于产生式分数空间的图像识别方法

摘要

针对传统图像识别方法没有利用数据分布信息的特点,提出了一种基于产生式分数空间的图像识别方法,使用概率产生式模型对数据分布建模,并从模型中学习出分数函数(特征映射),该分数函数编码了隐变量、样本和模型参数等用于分类的高层语义信息.最后,在两个标准图像数据库上对本文模型进行仿真和验证,实验结果验证了方法的有效性和稳定性.%Traditional image recognition approaches don't take use of information of data distribution.To overcome this,this paper proposes an image recognition method based on generative score space.Specifically,the approach models the data distribution using probabilistic generative model and derives the score function (i.e.feature mapping) from the model.The score function encodes high-level semantic information in hidden variables,sample and model parameters.At last,the proposed method is evaluated on two standard image datasets.The evaluation results validate the effectiveness and stability of our approach.