The LogitBoost Based on Joint Feature for Face Detection
Shishi Duan, Xiangyang Wang, Wanggen Wan · 2013
In this paper, joint Haar-like feature is used for detecting faces in images. Our method is based on the co-occurrence Haar-like features which can capture the structural characteristic of the face, make it possible to construct more effective weak classifier. As the Haar-like, the joint Haar-like feature can be calculated very effective and has the robustness to addition of noise and change in illumination. The face detector is learned by stage wise selection which is different with Viola and Jones Detector is that we use LogitBoost. We perform two experiments: In the Experiment 1, we show that the LogitBoost [9] obtain higher performance than AdaBoost [2] [4] [11]. We have confirmed that our method based on LogitBoost yielded higher performance than AdaBoost. In the Experiment 2, the performance increase according with the num of the combined features. However, the time spent increase at huge growth rate, too. Therefore, we will research that how we can choose the optimal F at the acceptable training time in the latter work.