Profile recognition based on co-training
Nobuhiko Mukai, Tomoki Shibamori, Youngha Chang · 2016
The technology of human face detection is widely used in cellular phones with cameras, and the technique is based on classification method using AdaBoost algorithm with Haar-like features [Viola and Jones 2001]. The method can detect frontal faces precisely; however, it is difficult to detect profiles with the same method. One of the reasons is that there is little training images of profile. On the other hand, some researches are using co-training [Sharma et al. 2008; Bhatt et al. 2011] to improve the precision of the classifiers. Therefore, we propose a profile recognition method based on co-training. The method constructs two kinds of similar profile classifiers, and the classifiers are re-trained with the images that have been detected by the other classifier. The method can use the experimental images as the training data so that the precision of the classifiers could be improved through the actual experiment even if they are trained with small number of data at first.