摘要
针对传统稀疏表示方法构建的字典不具备判别性的问题,以K-SVD算法为基础,对判别字典的构建和分类求解进行了研究,提出一种基于层次结构化字典学习的表情识别方法。先将训练样本切割出眼眉、脸颊和嘴三部分,对分割的各部分利用K-SVD算法得到块字典向量,再用层次分析法的权重赋值方法求块字典向量的权重值,构成各类子字典。将所有的子字典进行联合,用结构化字典学习算法求解。测试样本的归类取决于求解结果重构的效果。在JAFFE和CK表情库上的实验表明,该算法在保证了字典判别性的同时,也达到了较高的识别率。
Aiming at the constructed dictionary of traditional sparse representation method with not discrimination enough, this paper researched the construction and classification of discrimination dictionary based on K-SVD algorithm, and proposed a hierarchy structured dictionary learning method. It divided training samples into three parts, which were the eyes and eyebrows together, cheeks and mouth. It used K-SVD algorithm to obtain the block dictionary vectors for the divided three parts. It calculated the weight of block dictionary vectors by the weight assignment of AHP method, and then constructed sub-dictionary of each expression. At last, it combined all the constructed sub-dictionaries and used structured dictionary learning algorithm for solving. The classification of the test samples depended on the effect of the reconstruction. Experimental results show that the proposed method has guaranteed dictionary discrimination and also has a higher recognition rate by using JAFFE andCK database.
出处
《计算机应用研究》
CSCD
北大核心
2017年第11期3514-3517,共4页
Application Research of Computers
基金
广东省交通运输厅科技项目(科技-2016-02-030)
广东省自然科学基金博士启动项目(2014A030310169)
广东省自然科学基金面上项目(2016A030313703)
广东省自然科学基金资助项目(2016A030313713)
广东省科技计划资助项目(2016B030305002)
关键词
结构化字典
K-SVD算法
层次分析法
人脸表情识别
structured dictionary
K-SVD algorithm
analytic hierarchy process
expression recognition
