﻿ 基于分块CBP特征和稀疏表示的三维人脸表情识别
 计算机系统应用  2019, Vol. 28 Issue (2): 196-200 PDF

3D Facial Expression Recognition Using Block CBP Features and Sparse Representation
MA Jie, CAI Yi-Heng, SHENG Nan
Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
Foundation item: National Key Research and Development Program of China (SQ2017YFC170323); Science and Technology Program of Education Committee of Beijing Municipality (2017YFC1703302)
Abstract: In 3D facial expression recognition, operator algorithm based on LBP has the characteristics of accurate, precise, and invariable illumination, while it also has the disadvantages of high dimensions of histogram, poor discriminating ability, and large redundant information, comparing with traditional feature extraction method. In this study, a CPB algorithm is proposed based on multi-scaled block CBP feature extraction system for the classification of facial expressions that are represented in 3D, designating to extract features more efficiently. Then, sparse expression classifier is used to classify and identify the features. The experimental results demonstrate that the facial expression recognition rate has been greatly improved, comparing with traditional LBP algorithm and SVM algorithm.
Key words: LBP     CBP     feature extraction     sparse expression     feature block extraction

1 方法框架

 图 1 本文方法的算法流程图

1)面部多尺度分块. 为了更好地表达出原图像人脸表情的细节特征, 提高分类准确率, 本文首先对原始图像进行了不同尺度分块. 分块数量分别为1, 4, 6, 9, 16块, 如图2所示.

2) 局部CBP特征提取. 由于CBP特征对于图像的纹理特征具有较强的描述能力, 而这些特征是表情分类的重要依据, 本文基于前步分块结果, 提取各分块的局部CBP特征, 并计算CBP特征直方图作为分类特征向量.

3)稀疏表示分类. 同表情的人脸样本可看作一类, 其特征向量处于同一个线性空间, 任何一个测试样本都可以用来自于该类的训练样本进行线性表示; 将所有的训练样本特征向量构成字典, 则测试样本特征向量在该字典上的表示是稀疏的, 同时该稀疏系数包含了样本的类别信息. 基于此, 本文利用稀疏表示进行人脸识别.

 图 2 人脸区域划分图

1.1 分块CBP特征提取

 $\begin{split} CBP\left( {{\rm{8}},{\rm{1}}} \right) =& \sum\limits_{I = 0}^{\rm{3}} {{\rm{s}}({g_n} - {g_{n + 4}}){2^n} + s\left({g_c} - \frac{1}{9}\left(\sum\limits_{n = 0}^7 {{g_n} + {g_C}} \right)\right)} \;{2^4}\\ =& s({g_0} - {g_4}){2^0} + s({g_1} - {g_5}){2^1} + s({g_2} - {g_6}){2^2}\\ & + s({g_3} - {g_7}){2^3} + s({g_C} - {g_T}){2^4} \end{split}$ (1)

 $s(x) = \left\{ \begin{gathered} 1,\;\;\;\;\;x \geqslant |C| \\ 0,\;\;\;\;x < |C| \\ \end{gathered} \right.$ (2)
 图 3 不同表情的CBP特征图

1.2 稀疏表达分类器

 $y = DA$ (3)

 $A = [A_1^1,\cdots,A_m^1,A_1^2,\cdots,A_m^2,A_1^n,\cdots,A_m^n]$ (4)

 $\begin{gathered} \overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\frown}$}}{A} = \arg \min ||A|{|_1} \;\;\; {\rm{s.t.}} \;\;||DA - y|| < \varepsilon \\ \end{gathered}$ (5)

2 相关实验

 图 4 不同强度等级人脸表情图

(1)基于稀疏表示分类器的三维人脸表情识别准确率高于基于SVM分类器的识别效果. 从表中可以看出, 基于稀疏表示的三维人脸表情识别针对CBP特征与LBP特征的识别准确率分别为87.21%与75.90%, 均高于基于SVM的三维人脸表情识别针对CBP特征与LBP特征的85.78%与70.27%. 这说明稀疏表示分类方法对于三维人脸表情识别来说具有更好的分类识别能力. 提出的基于CBP特征与稀疏表示方法的三维人脸表情识别效果最高, 识别准确率达到了87.21%;

(2)基于CBP特征的三维人脸表情识别效果远优于基于LBP特征的识别效果. 表2中, 基于CBP特征的三维人脸表情识别准确率分别为85.78%和87.21%, 比基于LBP特征的识别准确率分别提高了15.51%和11.31%;

(3)本文所采用的基于CBP特征与稀疏表示分类的算法框架, 同Wang等人[13]和Berretti等人[14]的算法相比具有明显的优势, 分类识别的正确率分别高出3.6%和8.81. 实验1的特征提取对象是整个人脸表情图像, 人脸表情不同区域对于人脸表情的描述能力存在差异, 因此为了进一步验证本文提出的分块CBP特征的性能, 本文基于实验1中性能最好的CBP+稀疏表示算法框架, 进行了实验2的对比分析.

 图 5 不同分块数识别结果曲线

(1)随着人脸表情区域分块增加, 基于CBP特征与稀疏表示的三维人脸表情识别平均准确率呈增长趋势. 从表中可以看出, 基于1分块、4分块、6分块、9分块与16分块的三维人脸表情识别的平均准确率分别为87.2%、88.5%、89.3%、91.6%和89.8%, 三维人脸表情平均识别准确率先增加后减少, 其中9分块分类表现最佳.

(2)随着区域分块数目的增加, 三维人脸不同表情的识别准确率大多呈上升趋势. 从图中可以看出, 对于中性、高兴、愤怒、惊讶、悲伤与厌恶这6种人脸表情, 随着人脸分块数的增加, 识别准确率随着增加, 仅有恐惧表情例外, 这说明了人脸不同区域对于不同的表情影响程度不同, 合理地进行人脸区域分块能够有效地提升人脸表情识别准确率:

3 总结与展望

 [1] Chen JJ, Zhao Z, Sun H, et al. Facial expression recognition based on PCA reconstruction. 2010 5th International Conference on Computer Science & Education. Hefei, China. 2010. 195–198. [2] Edwards GJ, Taylor CJ, Cootes TF. Interpreting face images using active appearance model. Proceedings Third IEEE International Conference on Automatic Face and Gesture Recognition. Nara, Japan. 1998. 300–305. [3] Zhang HH, Zhang BL, Huang WM, et al. Gabor wavelet associative memory for face recognition. IEEE Transactions on Neural Networks, 2005, 16(1): 275-278. DOI:10.1109/TNN.2004.841811 [4] Ojala T, Pietikäinen M, Harwood D. A comparative study of texture measures with classification based on featured distributions. Pattern Recognition, 1996, 29(1): 51-59. DOI:10.1016/0031-3203(95)00067-4 [5] Heisele B, Serre T, Poggio T. A component-based framework for face detection and identification. International Journal of Computer Vision, 2007, 74(2): 167-181. DOI:10.1007/s11263-006-0006-z [6] Zaidi NA, Squire DM. Local adaptive SVM for object recognition. 2010 International Conference on Digital Image Computing: Techniques and Applications. Sydney, NSW, Australia. 2010. 196–201. [7] Rázuri JG, Sundgren D, Rahmani R, et al. Automatic emotion recognition through facial expression analysis in merged images based on an artificial neural network. 2013 12th Mexican International Conference on Artificial Intelligence. Mexico City, Mexico. 2013. 85–96. [8] Wright J, Yang AY, Ganesh A, et al. Robust face recognition via sparse representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009, 31(2): 210-227. DOI:10.1109/TPAMI.2008.79 [9] Mallat SG, Zhang ZF. Matching pursuits with time-frequency dictionaries. IEEE Transactions on Signal Processing, 1993, 41(12): 3397-3415. DOI:10.1109/78.258082 [10] Pati YC, Rezaiifar R, Krishnaprasad PS. Orthogonal matching pursuit: Recursive function approximation with applications to wavelet decomposition. Proceedings of 27th Asilomar Conference on Signals, Systems and Computers. Pacific Grove, CA, USA. 1993. 40–44. [11] Cai TT, Wang L. Orthogonal matching pursuit for sparse signal recovery with noise. IEEE Transactions on Information Theory, 2011, 57(7): 4680-4688. DOI:10.1109/TIT.2011.2146090 [12] Yin LJ, Wei XZ, Sun Y, et al. A 3D facial expression database for facial behavior research. 7th International Conference on Automatic Face and Gesture Recognition. Southampton, UK. 2006. 211–216. [13] Wang J, Yin LJ, Wei XZ, et al. 3D facial Expression recognition based on primitive surface feature distribution. 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. New York, NY, USA. 2006. 1399–1406. [14] Berretti S, Ben Amor B, Daoudi M, et al. 3D facial expression recognition using SIFT descriptors of automatically detected keypoints. The Visual Computer, 2011, 27: 1021-1036. DOI:10.1007/s00371-011-0611-x