Face detection based on AdaBoost algorithm with differential images
Hongjin Zhu, Shisong Zhu, Toshio Koga · 2008
Recently, a powerful face detection method based on AdaBoost algorithm is drawing attention to various applications. This method provides face detection systems with a good detection rate, although a considerable number of weak classifiers are needed. This paper introduces weak classifiers which can not be or can be less influenced by gradual brightness changes in face regions or changes in lighting condition. Using a simple mathematical model for these changes, we have found that a second-order differentiation, e.g. Laplacian Operator, is very useful to cope with these changes. In order to show the effectiveness, we have compared the classification results for original and differential images with and without normalization. As a result, the second-order differentiation is found to be very effective, regardless of normalization of images. This result suggests the number of weak classifiers may be reduced to a great extent, while preserving equal detection capability.