Multi-Stage Approach to Fast Face Detection

This paper describes a multi-stage approach for achieving fast and robust face detection. This approach was motivated by the work of Viola and Jones [7] using a cascade of classifiers yielding a coarse-to-fine strategy to significantly reduce detection time while maintaining high detection rate. However, it is distinguished from the previous work by two facts: (i) First, a new stage is added to more quickly estimate face candidate regions by using a larger window size and a larger moving step size. (ii) Second, we propose using SVM classifiers instead of AdaBoost classifiers in the last stage and study how to efficiently reuse Haar wavelet features selected by AdaBoost in the previous stage for SVM classifiers. By combining AdaBoost and SVM classifiers, the final system can obtain both fast and robust detection because most of the non-face patterns are quickly rejected in earlier layers, while only a small number of promising face patterns are robustly classified in the later layers. Extensive experimental results demonstrated that our proposed system can achieve promising results.