摘要
针对在较大的人脸平面旋转(RIP)角度下,人脸外观差异导致多尺度旋转人脸检测困难的问题,提出了一种基于级联回归网络的多尺度旋转人脸检测方法,该方法采用三级级联回归策略逐步完成每个候选人脸RIP角度的估计和人脸候选窗位置的更新。第1阶段采用SSD网络对人脸候选窗位置进行直接预测,消除大部分低置信度的人脸候选窗;第2阶段采用人脸分类网络对人脸候选窗位置进行粗略回归,同时快速对齐人脸的RIP主方向区间;第3阶段采用人脸回归网络对人脸候选窗位置进行精确回归,同时连续估计人脸RIP的角度。在旋转FDDB数据集和旋转WIDER FACE数据集上的实验结果表明,该算法对多尺度旋转人脸的检测精度高于当前主流的旋转人脸检测算法。
Aiming at the difficulty of multi-scale rotating face detection caused by the significant difference of face feature,under the arbitrary rotation-in-plane( RIP) angles,this paper proposes a multi-scale rotating face detection method based on cascade regression network. A three-level cascade regression strategy is adopted to gradually complete the estimation of face RIP angle and update of face window position of each candidate. Firstly,the SSD network is used to directly predict the face candidate window position,and most of the low-confidence face candidate windows are eliminated. Secondly,the face classification network is used to make a rough regression of the face candidate window position,and at the same time,the RIP main direction range of the face is quickly aligned. Finally,the face regression network is used to accurately regress the face candidate window position,and the angle of the face RIP is continuously estimated. The experiments on the rotating FDDB dataset and the rotating WIDER FACE dataset show that the detection accuracy of the multi-scale rotating face are better than those of the current mainstream rotating face detection algorithm.
引文
[1]FARFADE S S,SABERIAN M J,LI L J.Multi-view face detection using deep convolutional neural networks[C].ACM on International Conference on Multimedia Retrieval,2015:643-650.
[2]LI H,LIN Z,SHEN X,et al.A convolutional neural network cascade for face detection[C].Computer Vision and Pattern Recognition,IEEE,2015:5325-5334.
[3]LIU W,ANGUELOV D,ERHAN D,et al.SSD:Single shot MultiBox detector[C].European Conference on Computer Vision.Springer International Publishing,2016:21-37.
[4]NAJIBI M,SAMANGOUEI P,CHELLAPPA R,et al.SSH:Single stage headless face detector[C].International Conference on Computer Vision,IEEE,2017:4885-4894.
[5]HU P,RAMANAN D.Finding tiny faces[C].Computer Vision and Pattern Recognition,IEEE,2017:1522-1530.
[6]JIANG H,LEARNED-MILLER E.Face Detection with the faster R-CNN[C].International Conference on Automatic Face Gesture Recognition,IEEE,2016:650-657.
[7]HAO Z,LIU Y,QIN H,et al.Scale-aware face detection[C].Computer Vision and Pattern Recognition,IEEE,2017:1913-1922.
[8]KAN M,KAN M,SHAN S,et al.Funnel-structured cascade for multi-view face detection with alignmentawareness[J].Neurocomputing,2017,221(C):138-145.
[9]PRIYA G N,BANU R S D W.A Robust rotation invariant multiview face detection in erratic illumination condition[J].International Journal of Computer Applications,2012,57(20):46-51.
[10]KYLBERG G,SINTORN I M.On the influence of interpolation method on rotation invariance in texture recognition[J].Eurasip Journal on Image and Video Processing,2016,2016(1):1-12.
[11]DU S,LIU J,LIU Y,et al.Precise glasses detection algorithm for face with in-plane rotation[J].Multimedia Systems,2017,23(3):293-302.
[12]SHI X,SHAN S,KAN M,et al.Real-time rotationinvariant face detection with progressive calibration networks[C].Computer Vision and Pattern Recognition,IEEE,2018.
[13]LI H,LIN Z,SHEN X,et al.A convolutional neural network cascade for face detection[C].Computer Vision and Pattern Recognition,IEEE,2015:5325-5334.
[14]JAIN V,LEARNED-MILLER E.FDDB:A benchmark for face detection in unconstrained settings[R].UMass Amherst Technical Report.2010.
[15]YANG S,LUO P,CHEN C L,et al.WIDER FACE:Aface detection benchmark[C].IEEE Conference on Computer Vision and Pattern Recognition,2016:5525-5533.
[16]REN S,HE K,GIRSHICK R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[C].International Conference on Neural Information Processing Systems,MIT Press,2015:91-99.