搜集整理了2004~2015性能最好的人脸检测的部分资料,欢迎交流和补充相关资料。
1:人脸检测性能
1.1 人脸检测测评
目前有两个比较大的人脸测评网站:
1:Face Detection Data Set and Benchmark(FDDB)
网址:http://vis-www.cs.umass.edu/fddb/results.html
FDDB是由马萨诸塞大学计算机系维护的一套公开数据库,为来自全世界的研究者提供一个标准的人脸检测评测平台,其中涵盖在自然环境下的各种姿态的人脸;该校还维护了LFW等知名人脸数据库供研究者做人脸识别的研究。作为全世界最具权威的人脸检测评测平台之一,FDDB使用Faces in the Wild数据库中的包含5171张人脸的2845张图片作为测试集,而其公布的评测集也代表了人脸检测的世界最高水平。
FDDB更新更及时一些,所以本文的资料还是主要参考的FDDB。
2:Fine-grained evaluation of face detection in the wild
网址:http://www.cbsr.ia.ac.cn/faceEvaluation/results.html
该测试网站是由李子青老师的研究组创立和维护的,其性能评估更细致,分析不同分辨率、角度、性别、年龄等条件下的算法准确率。该测试集更新没有FDDB及时。
1.2 Suevey
1)2010年微软zhang cha和张正友撰写的人脸检测的综述报告
[MSR-TR-2010] A_survey_of_recent_advances_in_face_detection
2)Stefanos Zafeiriou, Cha Zhang和张正友撰写了最新的人脸检测的综述paper,将出版在2016年的《Computer Vision and Image Understanding》
[CVIU 2015] A Survey on Face Detection in the wild past, present and future
最新性能总结如下:
1)在过去的10年人脸检测的性能已经有了激动人心的提升。
2)这些引人注目的性能提升,主要还是得益于将Viala-Jones的boosting和鲁棒性的特征相组合。
3)始终有15~20%的性能Gap,即使允许一个相对较大的FP(大约1000),始终有15~10%的人脸无法被检测到。需要特别指出的是这些Gap主要是由于是失焦的人脸(比如模糊的人脸)。
4)在这个Benchmark中,最好的基于boosting技术和最好的基于DPM的技术是比较接近的。当然最好的技术还是boosting和DPM组合在一起的性能。(这个就是指的[ECCV 2014] Joint Cascade Face Detection and Alignment)
1.4 有关人脸检测指标
如果对于人脸检测指标不是很熟悉,可以参考http://www.cvrobot.net/recall-precision-false-positive-false-negative/
2. 2014的进展
1:Joint Cascade Face Detection and Alignment. ECCV 2014. D. Chen, S. Ren, Y. Wei, X. Cao, J. Sun.
Paper:[ECCV 2014] Joint Cascade Face Detection and Alignment
中文介绍:联合人脸检测、校准算法介绍
2:The fastest deformable part model for object detection J. Yan, Z. Lei, L. Wen, S. Z. Li,
paper:[CVPR 2014] The Fastest Deformable Part Model for Object Detection
3:Face detection without bells and whistles. ECCV 2014. M. Mathias, R. Benenson, M. Pedersoli and L. Van Gool.
Paper:[ECCV 2014] Face detection without bells and whistles.
project:http://markusmathias.bitbucket.org/2014_eccv_face_detection/
Code:https://bitbucket.org/rodrigob/doppia
Talk: http://videolectures.net/eccv2014_mathias_face_detection/ (不错的报告)
Slide:eccv2014_mathias_face_detection_01
4:A Method for Object Detection Based on Pixel Intensity Comparisons Organized in Decision Trees. CoRR 2014. N. Markus, M. Frljak, I. S. Pandzic, J. Ahlberg and R. Forchheimer.
Code:https://github.com/nenadmarkus/pico
Paper:Object Detection with Pixel Intensity Comparisons Organized in Decision Trees
实时人脸检测视频Demo:
5:Aggregate channel features for multi-view face detection.. B. Yang, J. Yan, Z. Lei and S. Z. Li.
Paper:[IJCB 2014] Aggregate channel features for multi-view face detection
3. 2015的最新进展
6:A Convolutional Neural Network Cascade for Face Detection. H. Li , Z. Lin , X. Shen, J. Brandt and G. Hua.
paper:[CVPR2015] A Convolutional Neural Network Cascade for Face Detection
7:Multi-view Face Detection Using Deep Convolutional Neural Networks. S. S. Farfade, Md. Saberian and Li-Jia Li
这是yahoo的人脸检测
Paper:[ICMR 2015] Multi-view Face Detection Using Deep Convolutional Neural Networks
News:The Face Detection Algorithm Set to Revolutionize Image Search
8:Face Detection with a 3D Model. A. Barbu, N. Lay, G. Gramajo.
paper:Face Detection with a 3D Model
结论:The 3D proposals are not perfectly aligned with the face keypoints, which results in a reduced accuracy in the high precision/very low false positive regime compared to other state of the art methods. However, in the regime of at least 0.1 false positives per image, it outperforms the cascade-based state of the art methods.