摘要： 人脸语义检索在识别技术中有着重要的作用，如表情检索、性别判断、年龄估计等，上述识别技术通过捉捕人脸语义信息来实现。研究将人脸语义信息融入到人脸检索中，提出一种基于稀疏学习的人脸语义子空间提取方法。语义子空间学习被分为字典构建和稀疏学习2个部分。在字典构建的过程中，给出语义差的方法来对互斥语义进行计算，使提取的某类语义不受其他类语义干扰语义子空间，并对不同语义环境和不同语义差组合进行测试。在稀疏学习部分，使用Lasso算法对其进行改进。实验结果表明，与传统Fisher方法相比，该方法撇除其他语义干扰的子空间稳定性更强，且有一定的降维效果。

关键词: 子空间学习, 人脸语义, 稀疏学习, 人脸识别

Abstract: Face semantics retrieval is a key point in today’s biometric recognition technology. Such as facial expressions recognition, gender classification and age estimation, they all accomplish their functions by catching semantics. This paper researches face semantic information in face retrieval, and proposes a face semantic subspace extraction method. Semantic subspace learning is divided into dictionary building and sparse learning. In the process of dictionary building, this paper gives the method of semantic difference to calculate mutually exclusive semantics, and extracted semantics is not disturbed by other semantics. Through testing different combination in different semantic environment, result proves that the method is more stable. In sparse learning, Lasso algorithm is improved, and result shows that compared with Fisher method, the subspace effect has increasement.

Key words: subspace learning, face sematic, sparse learning, face recognition

中图分类号: 

• TP18