Component-based face detection method for various types of occluded faces
Kiyoto Ichikawa, Takeshi Mita, Osamu Hori, Takao Kobayashi · 2008
This paper proposes a method that can be used to detect various types of occluded faces as well as non-occluded faces by using classifiers based on AdaBoost, linear discriminant analysis (LDA), and a decision tree structure. The proposed method involves AdaBoost-based classifiers for whole faces and individual face-part classifiers trained on non-occluded face sample sets. Whole faces and their parts are classified individually and the final decision is made by combining the outputs from all the classifiers. We used a combination of a decision tree trained by the C4.5 algorithm and LDA to combine the outputs. The decision tree is designed to classify non-occluded faces and various types of occluded faces. The experimental results revealed that the proposed method was extremely effective in detecting both non- occluded and various types of occluded faces.