基于稀疏表示和机器学习的人脸识别系统研究

发布时间：2018-08-07 07:11

【摘要】：人脸识别是机器视觉和模式识别领域最富有挑战性的课题之一,同时也具有较为广泛的应用意义,在电子商务、公安系统、档案管理、人机交互等各个领域中都有很高的应用与商业价值。但目前的人脸识别系统在实际应用中仍然存在不少的缺陷:如在光照、环境、姿态的影响下,识别率还不尽如人意。因此,本论文研究了一种结合机器学习与稀疏表示的人脸识别系统,使其能够在噪声环境下有较好的识别率。论文的主要研究内容与成果如下:(1)首先针对人脸识别在噪声环境下表现不够良好的问题,提出了基于稀疏表示的分类器算法来进行识别,从而提高了在人脸有遮挡或者光照条件不佳下的识别率。(2)由于传统人脸特征提取方式的欠缺,导致提取到的特征过于简单与初级。针对此问题,提出了基于深度学习理论的级联卷积自编码器来提取特征。级联的卷积自编码器由于其能够进行自我学习,减少特征提取的误差,并且通过多层级联可以提取更为深层和抽象的特征,进一步提高了系统的识别率。(3)结合了两种方式:基于稀疏表示的分类器算法以及基于深度学习理论的级联卷积自编码器,设计并实现了人脸识别系统。此系统在噪声环境下具有良好的表现,并且识别速度快,准确率高。
[Abstract]:Face recognition is one of the most challenging topics in the field of machine vision and pattern recognition. Human-computer interaction and other fields have high application and commercial value. However, the current face recognition system still has many defects in practical application: for example, under the influence of illumination, environment and attitude, the recognition rate is not satisfactory. Therefore, a face recognition system based on machine learning and sparse representation is proposed in this paper, which can achieve better recognition rate in noisy environment. The main research contents and results are as follows: (1) aiming at the problem that face recognition does not perform well in noise environment, a sparse representation based classifier algorithm is proposed. Therefore, the recognition rate of face with occlusion or poor illumination is improved. (2) because of the lack of traditional face feature extraction, the extracted features are too simple and primary. To solve this problem, a concatenated convolution self-encoder based on depth learning theory is proposed to extract features. The concatenated convolution self-encoder can learn from itself, reduce the error of feature extraction, and extract deeper and abstract features by multi-layer cascade. The recognition rate of the system is further improved. (3) A face recognition system is designed and implemented by combining two approaches: a sparse representation based classifier algorithm and a cascaded convolution self-coder based on depth learning theory. The system has good performance in noise environment and high recognition speed and accuracy.
【学位授予单位】：浙江工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP181

【相似文献】