Multi-view Face Detection Based on the Enhanced AdaBoost Using Walsh Features
Yunyang Yan, Zhibo Guo, Jingyu Yang · 2007
A novel face detection algorithm is proposed in this paper to improve the training speed and detection performance. Firstly, we used Walsh features instead of Haar-like features in the AdaBoost algorithm. Walsh features have less redundancy than Haar-like features due to its orthogonal specialty. Then, we defined a kind of week classifiers with dual-threshold to speedup training process and increase accuracy. Furthermore, during training, dual-threshold of every classifier is adoptively adjusted to separate the face and non-face as far as possible. Experimental results on MIT+CMU frontal face set and CMU profile face set demonstrated that the proposed technique can achieve better results on the detection speed and accuracy than the corresponding method.