基于单演二值编码与稀疏编码人脸识别算法的研究

发布时间：2018-06-18 10:17

  本文选题：人脸识别 + 单演二值编码　； 参考：《湖南师范大学》2016年硕士论文

【摘要】：人脸识别在机器学习与数字图像处理领域是一个非常受欢迎但又极具有挑战性的课题。因为它有许多优点,比如非接触性、隐蔽性、以及图像采集设备成本低等,已经被越来越多地应用于信息安全、人机交互、人工智能以及电子商务等安全中。但在实际应用中,许多非可控条件比如遮挡、姿态变化、光照、表情和时间等制约人脸识别的性能,特别是条件改变厉害的时候,性能较好的识别系统识别率也会快速下降。文章从特征提取、数据降维和分类识别三个方面对人脸识别算法展开广泛的研究,论文的主要工作包括:(1)对人脸识别这一问题进行了简单的描述,综合分析和概括了人脸识别技术目前的发展现状。(2)总结了已有的全局特征提取与局部特征提取算法,全局特征对姿态变化和遮挡非常敏感,一般用来进行粗略的匹配,而局部特征一般为人脸识别提供细致的确认。(3)研究了基于单演二值编码的人脸识别算法,良好的人脸特征提取算法是鲁棒高效的人脸识别算法能否成功的关键因素,融合局部幅值、相位以及方向的单演二值编码是目前最好的特征提取方法之一。(4)研究了基于稀疏编码的人脸识别算法,稀疏编码算法有多种模型,主要研究了基于稀疏表示与基于协同表示的人脸识别算法,基于稀疏表示的人脸识别算法是以最小的重构误差来进行分类的,但是稀疏表示的计算代价非常高。基于协同表示的人脸识别算法是稀疏编码的几种模型中速度最快的之一。(5)尽管基于稀疏表示的人脸识别算法非常新颖有效,但是有一个问题需要进一步解决。用来测试的特征脸、随机脸和Fisher脸都是全局特征,因为在实际应用中,训练样本通常都是受到限制的,这样的全局特征不能有效地处理光照、表情和姿势等变化。为此,提出一种新颖的融合单演二值编码与稀疏编码的人脸识别算法,在提取人脸局部特征之后,采用稀疏表示的方法降维,考虑到算法的识别率与算法运行时间,采用稀疏编码模型中的协同表示模型。在ORL、AR和PolyU-NIR人脸库上测试,实验结果表明,该算法相较于传统的稀疏编码算法,性能有所改善。
[Abstract]:Face recognition is a very popular and challenging topic in the field of machine learning and digital image processing. Because it has many advantages, such as non-contact, concealment, and low cost of image acquisition equipment, it has been increasingly used in information security, human-computer interaction, artificial intelligence and e-commerce security. However, in practical applications, many uncontrollable conditions such as occlusion, attitude change, illumination, facial expression and time restrict the performance of face recognition, especially when the conditions change severely, the recognition rate of the recognition system with better performance will decline rapidly. In this paper, face recognition algorithms are widely studied from three aspects: feature extraction, data reduction and classification recognition. The main work of this paper includes: 1) briefly describing the problem of face recognition. The present development of face recognition technology is analyzed and summarized. The existing algorithms of global feature extraction and local feature extraction are summarized. The global feature is very sensitive to the change of posture and occlusion, and is generally used for rough matching. However, local features generally provide detailed recognition for face recognition. (3) A face recognition algorithm based on single binary coding is studied. A good face feature extraction algorithm is a key factor to the success of robust and efficient face recognition algorithm. Fusion of local amplitude, phase and direction is one of the best feature extraction methods. (4) face recognition algorithm based on sparse coding is studied. There are many models in sparse coding algorithm. Face recognition algorithms based on sparse representation and cooperative representation are mainly studied. The face recognition algorithm based on sparse representation is classified with minimal reconstruction error, but the computational cost of sparse representation is very high. The face recognition algorithm based on cooperative representation is one of the fastest among several models of sparse coding. Although the face recognition algorithm based on sparse representation is novel and effective, there is one problem that needs to be solved further. The feature faces, random faces and Fisher faces used for testing are all global features, because in practical applications, the training samples are usually restricted, so the global features can not effectively deal with the changes of illumination, expression and posture. In this paper, a novel face recognition algorithm combining single binary coding and sparse coding is proposed. After extracting the local features of the face, the method of sparse representation is used to reduce the dimension, considering the recognition rate of the algorithm and the running time of the algorithm. The cooperative representation model in sparse coding model is adopted. The experimental results on ORLPAR and PolyU-NIR face databases show that the performance of the proposed algorithm is better than that of the traditional sparse coding algorithm.
【学位授予单位】：湖南师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
，

本文编号：2035134