基于自适应特征的遮挡人脸特征点定位算法
|
摘要
为了使人脸特征点定位算法在人脸被物体遮挡的情况下仍能快速、准确地检测特征点的位置,提出了一种基于自适应特征的遮挡人脸特征点定位算法。该方法首先检测每个特征点的遮挡状态,即先训练一个逻辑回归模型,通过所有特征点周围的纹理特征快速地估计出每个特征点被遮挡的概率值;然后根据每一个特征点被遮挡的概率自适应地调整该特征点纹理特征的权重,使得被遮挡概率较大的特征点获得较小的权重值,减小人脸遮挡对特征的影响,提高特征点定位的准确度。实验结果表明,本文算法的特征点定位的平均误差达到5. 94%,遮挡检测准确率/召回率达到80%/72. 84%。
In order to make the facial landmark localization algorithm detect the position of feature points quickly and accurately when the face is occluded by objects, this paper proposes a facial landmark localization under occlusion method based on adaptive feature extraction. The method first detected the occlusion state of each feature point, that is, first trained a logistic regression model, and quickly estimated the probability value of each feature point being occluded with the texture features around all landmarks. Then, adaptive weights were assigned to each feature point according to their estimated occlusion probability, so that the feature points with larger occlusion probabilities were assigned smaller weight values, therefore the impact of facial occlusion on the feature was decreased, and the accuracy of landmark localization was improved. Quantitative experiments on the challenging COFW benchmark show that the proposed method obtains the state-of-art results in terms of localization accuracy and occlusion detection, with an average localization error of 5.94% and a precision/recall of 80%/72.84%.
In order to make the facial landmark localization algorithm detect the position of feature points quickly and accurately when the face is occluded by objects, this paper proposes a facial landmark localization under occlusion method based on adaptive feature extraction. The method first detected the occlusion state of each feature point, that is, first trained a logistic regression model, and quickly estimated the probability value of each feature point being occluded with the texture features around all landmarks. Then, adaptive weights were assigned to each feature point according to their estimated occlusion probability, so that the feature points with larger occlusion probabilities were assigned smaller weight values, therefore the impact of facial occlusion on the feature was decreased, and the accuracy of landmark localization was improved. Quantitative experiments on the challenging COFW benchmark show that the proposed method obtains the state-of-art results in terms of localization accuracy and occlusion detection, with an average localization error of 5.94% and a precision/recall of 80%/72.84%.
引文
[1]王燕,王双印.基于卷积神经网络的人脸信息增强识别研究[J].计算机科学,2018,45(8):268-271.DOI:10.11896/j.issn.1002-137X.2018.08.048.WANG Y,WANG S Y.Research on face information enhancement and recognition based on convolutional neural network[J].Computer Science,2018,45(8):268-271.DOI:10.11896/j.issn.1002-137X.2018.08.048(Ch).
[2]DING C X,TAO D C.Pose-invariant face recognition with homography-based normalization[J].Pattern Recognition,2017,66:144-152.
[3]苏志铭,陈靓影.基于自回归模型的动态表情识别[J].计算机辅助设计与图形学学报,2017,29(6):1085-1092.SU Z M,CHEN L Y.An auto-regressive model based approach to dynamic facial expression recognition[J].Journal of Computer-Aided Design&Computer Graphics,2017,29(6):1085-1092(Ch).
[4]BURGOS-ARTIZZU X P,PERONA P,DOLLAR P.Robust Face landmark estimation under occlusion[C]//IEEE International Conference on Computer Vision.Washington D C:IEEE.2013:1513-1520.DOI:10.1109/ICCV.2013.191.
[5]LIU Q S,DENG J K,TAO D C.Dual sparse constrained cascade regression for robust face alignment[J].IEEE Transactions on Image Processing,2016,25(2):700-712.DOI:10.1109/TIP.2015.2502485.
[6]LIU Q S,DENG J K,YANG J,et al.Adaptive cascade regression model for robust face alignment[J].IEEETransactions on Image Processing,2017,26(2):797-807.DOI:10.1109/TIP.2016.2633939.
[7]YU X,LIN Z,BRANDT J,et al.Consensus of regression for occlusion-robust facial feature localization[C]//European Conference on Computer Vision.Cham:Springer.2014:105:118.
[8]刘袁缘,谢忠,周顺,等.基于条件迭代更新随机森林的非约束人脸特征点精确定位[J].计算机辅助设计与图形学学报,2017,29(10):1881-1890.DOI:10.3969/j.issn.1003-9775.2017.10.014.LIU Y Y,XIE Z,ZHOU S,et al.Conditional iteration updated random forests for unconstrained facial feature location[J].Journal of Computer-Aided Design&Computer Graphics,2017,29(10):1881-1890.DOI:10.3969/j.issn.1003-9775.2017.10.014(Ch).
[9]WU Y,GOU C,JI Q.Simultaneous facial landmark detection,pose and deformation estimation under facial occlusion[C]//IEEE Conference on Computer Vision and Pattern Recognition.Washington D C:IEEE,2017:5719-5728.
[10]DALAL N,TRIGGS B.Histograms of oriented gradients for human detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Washington D C:IEEE,2005:886-893.DOI:10.1109/CVPR.2005.177.
[11]VALLE R,BUENAPOSADA J M,VALDES A,et al.A deeply-initialized coarse-to-fine ensemble of regression trees for face alignment[C]//European Conference on Computer Vision.Berlin:Springer,2018:585-601.
[12]XING J L,NIU Z H,HUANG J S,et al.Towards robust and accurate multi-view and partially-occluded face alignment[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(4):987-1001.DOI:10.1109/TPAMI.2017.2697958.
[13]XIONG X H,DE LA TORRE F.Supervised descent method and its applications to face alignment[C]//IEEEConference on Computer Vision and Pattern Recognition.Washington D C:IEEE.2013:532-539.DOI:10.1109/CVPR.2013.75.
[14]FENG Z H,HUBER P,KITTLER J,et al.Random cascaded-regression copse for robust facial landmark detection[J].IEEE Signal Processing Letters,2015,22(1):76-80.DOI:10.1109/LSP.2014.2347011.
[15]SHAO Z W,ZHU H L,HAO Y Y,et al.Learning a multi-center convolutional network for unconstrained face alignment[C]//IEEE International Conference on Multimedia and Expo(ICME).Washington D C:IEEE,2017:114.DOI:10.1109/ICME.2017.8019505.
[2]DING C X,TAO D C.Pose-invariant face recognition with homography-based normalization[J].Pattern Recognition,2017,66:144-152.
[3]苏志铭,陈靓影.基于自回归模型的动态表情识别[J].计算机辅助设计与图形学学报,2017,29(6):1085-1092.SU Z M,CHEN L Y.An auto-regressive model based approach to dynamic facial expression recognition[J].Journal of Computer-Aided Design&Computer Graphics,2017,29(6):1085-1092(Ch).
[4]BURGOS-ARTIZZU X P,PERONA P,DOLLAR P.Robust Face landmark estimation under occlusion[C]//IEEE International Conference on Computer Vision.Washington D C:IEEE.2013:1513-1520.DOI:10.1109/ICCV.2013.191.
[5]LIU Q S,DENG J K,TAO D C.Dual sparse constrained cascade regression for robust face alignment[J].IEEE Transactions on Image Processing,2016,25(2):700-712.DOI:10.1109/TIP.2015.2502485.
[6]LIU Q S,DENG J K,YANG J,et al.Adaptive cascade regression model for robust face alignment[J].IEEETransactions on Image Processing,2017,26(2):797-807.DOI:10.1109/TIP.2016.2633939.
[7]YU X,LIN Z,BRANDT J,et al.Consensus of regression for occlusion-robust facial feature localization[C]//European Conference on Computer Vision.Cham:Springer.2014:105:118.
[8]刘袁缘,谢忠,周顺,等.基于条件迭代更新随机森林的非约束人脸特征点精确定位[J].计算机辅助设计与图形学学报,2017,29(10):1881-1890.DOI:10.3969/j.issn.1003-9775.2017.10.014.LIU Y Y,XIE Z,ZHOU S,et al.Conditional iteration updated random forests for unconstrained facial feature location[J].Journal of Computer-Aided Design&Computer Graphics,2017,29(10):1881-1890.DOI:10.3969/j.issn.1003-9775.2017.10.014(Ch).
[9]WU Y,GOU C,JI Q.Simultaneous facial landmark detection,pose and deformation estimation under facial occlusion[C]//IEEE Conference on Computer Vision and Pattern Recognition.Washington D C:IEEE,2017:5719-5728.
[10]DALAL N,TRIGGS B.Histograms of oriented gradients for human detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Washington D C:IEEE,2005:886-893.DOI:10.1109/CVPR.2005.177.
[11]VALLE R,BUENAPOSADA J M,VALDES A,et al.A deeply-initialized coarse-to-fine ensemble of regression trees for face alignment[C]//European Conference on Computer Vision.Berlin:Springer,2018:585-601.
[12]XING J L,NIU Z H,HUANG J S,et al.Towards robust and accurate multi-view and partially-occluded face alignment[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2018,40(4):987-1001.DOI:10.1109/TPAMI.2017.2697958.
[13]XIONG X H,DE LA TORRE F.Supervised descent method and its applications to face alignment[C]//IEEEConference on Computer Vision and Pattern Recognition.Washington D C:IEEE.2013:532-539.DOI:10.1109/CVPR.2013.75.
[14]FENG Z H,HUBER P,KITTLER J,et al.Random cascaded-regression copse for robust facial landmark detection[J].IEEE Signal Processing Letters,2015,22(1):76-80.DOI:10.1109/LSP.2014.2347011.
[15]SHAO Z W,ZHU H L,HAO Y Y,et al.Learning a multi-center convolutional network for unconstrained face alignment[C]//IEEE International Conference on Multimedia and Expo(ICME).Washington D C:IEEE,2017:114.DOI:10.1109/ICME.2017.8019505.