基于概率图模型的表情分类方法研究

Research on expression classification method based on probability graph model

针对在小规模样本集上如何提高学习算法的准确率问题,提出了基于概率图模型的表情分类算法.本文提出了一种表情区域分割方法,将人脸表情图像划分为5个面部区域,通过概率图模型的分类方法理论基础,由5个表情分类子网络和Softmax分类层构成基于概率图模型的表情分类模型,实现对人脸表情图像的分类.通过在JAFFE人脸表情库和CK表情数据库上实验分析,得到识别准确率分别为97.78%和98.95%,分别提高了1.85%和5.92%准确率.实验结果表明:本文方法对表情分类识别率的提高有重要意义,并且本文方法有效提高了对于小样本图像的分析与理解能力.

Aming at on how to improve the accuracy of learning algorithms on small-scale sample sets, an expression classification algorithm based on probability graph model is proposed. This paper proposes an expression region segmentation method, which divides the facial expression image into five face regions. Based on the theoretical basis of the classification method of the probability map model, the expressions based on the probability map model are composed of five expression classification sub-networks and Softmax classification layers.Classification model to achieve classification of facial expression images. Through the experimental analysis on the JAFFE facial expression database and the CK expression database, this study obtained recognition accuracy of97.78% and 98.95%, which improved the accuracy by 1.85% and 5.92% respectively. The experimental results show that the proposed method has important significance for improving the recognition rate of expressions, and the method effectively improves the analysis and understanding ability of small sample images.