| 吴迪,唐勇奇,林国汉,胡慧.基于显著性局部定向模式和深度学习的鲁棒人脸识别[J].光电子激光,2016,27(6):655~661 |
| 基于显著性局部定向模式和深度学习的鲁棒人脸识别 |
| Roboust face recognition based on significance local directional pattern and dee p learning |
| 投稿时间：2015-10-07 |
| DOI： |
| 中文关键词: 计算机应用 局部定向模式(LDP) 深度学习(DL) 显著性 人脸识别 |
| 英文关键词:computer application local directional pattern (LDP) deep learning significan ce face recognition |
| 基金项目:国家科技支撑计划(1214ZGA008)、国家自然科学基金(61263031,0)、湖南省自然科 学基金(14JJ6041,16JJ6025)、湖南省教育厅项目(14k029,5A044)、重庆市教委自然科学基金项目(KJ 1400628)和湖南工程学院博士科研启动基金(15045) (湖南工程学院 电气信息学院,湖南 湘潭 411004) |
|
| 摘要点击次数: 670 |
| 全文下载次数: 0 |
| 中文摘要: |
| 针对传统人脸识别算法特征提取不充分和对噪声敏 感的问题,提出了一种基于显著性局部定向模式(SLDP)和深 度学习(DL)的人脸识别算法。首先基于信息的显著性提出一种改进的SLDP人脸描述方法,在 利用局部井型 领域梯度信息的基础上,使用两组梯度值中各自最大值和次大值的方向编码成二位八进制数 ,从而形成改进的 SLDP编码,通过统计选出最具有显著性的SLDP编码作为特征向量；然后将改进的SLDP编码作 为深度神经网 络(DBN)的输入,通过逐层贪婪训练网络,获得良好的网络参数,并用训练好的网络对测试 样本进行预测。仿真实验表明, 本文所提出的人脸识别算法相对于传统识别算法识别率高,并且在对抗噪声方面比其它算法 更具有鲁棒性。 |
| 英文摘要: |
| In order to solve the problem of insuf ficient featu re extraction and sensitive to the noise of tranditional face recognition methods,a new face recotnition algorithm based on significance loca l directional pattern (SLDP) and deep learning (DL) is proposed in this paper.First,a new face feature extraction method based on SLDP is given,the local # shape neighborhood of each pixel is equally divided into 3×3sub-neigh borhoods,then eight edge gradient values of each sub-neighborhood are obtained by convolving it with eight Kirsch masks respectively,and SLDP just utilizes the directions of the most prominent edge response value and the second most prominent one.The SLDPs are generated by statistical analysis according to the occurrence rates of the SLDP codes in a ma ss of facial images.Finally,the face descriptor is represented by using the global concatenated histogram based on SLDP extracte d from the face image which is divided into several sub-regions.Second,making SLDP texture features as the input of d eep belief net (DBN),and greedy training for the network layer is made layer by layer to obtain good network parameters.At last,t he trained net is used to predict the test samples labels.The simulation results show that the algorithm can obtain higher recognition rate than traditional algorithms. |
| 查看全文 下载PDF阅读器 |
| 关闭 |
|
|
|
|
|