Analysis of the work of the front classifiers of the OpenCV library.
Abstract
The article analyzes five models of classifiers for detecting frontal faces of the standard OpenCV 3.0.0 package. and the features of their practical use are determined. On the basis of the analysis of the features of cascade classifiers, a computational experiment was performed to determine the accuracy, completeness and performance of each classifier. As input models, FDDB and AFW data sets are used that allow us to evaluate the work of natural person search algorithms. To evaluate the algorithms, a cross-check on ten subsets of images was used, with subsequent averaging of the results. In order to analyze the correlation between levels of true and false detections and to evaluate the quality of work for each classifier, PR and ROC curves were constructed. The calculations performed showed that the OpenCV-lbp classifier showed the highest performance, however, the most effective for detecting frontal faces according to all the parameters considered is the use of the OpenCV-alt classifier.
References
Baggio D.L. Mastering OpenCV with Practical Computer Vision Projects/ Baggio D. L., Emami S., Escrivá D.M. : Packt Publishing. – 2012. – 321 p.
Laganière R. OpenCV 2 Computer Vision Application Programming Cookbook/ R.Laganière : Packt Publishing. – 2011. – 298 р.
OpenCV. [Електронний ресурс]. – Режим доступу: http://opencv.org/
Lienhart R, Maydt J. An extended set of Haar-like features for rapid object detection. IEEE International Conference on Image Processing; 2002: 1: 900 – 903.
Freund Yoav. A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting / Yoav Freund, Robert E. Schapire. // Journal of computer and system sciences, 1997 – 55. – p. 119 –139.
Kalinovskii I.A. Review and testing of frontal face detectors. / I.A. Kalinovskii, V.G. Spitsyn. // Computer Optics. 2016 – 40(1) – p. 99– 111.
Viola P. Robust real-time face detection/ P.Viola, M. Jones, // IJCV – 2004. – 57(2), – p 137–154.
Davis J. The relationship between Precision-Recall and ROC curves / J. Davis, M. Goadrich // International Conference on Machine Learning. – 2006. – p. 233 – 240.
Abstract views: 76 PDF Downloads: 87
