Segment-Boost Learning for Facial Feature Selection

Recently, a lot of boosting-based feature selection methods have been introduced. However, the effective methods that can select more discriminative features at lower computational cost are needed. In this paper, we propose a novel boosting algorithm, called the Segment-Boost, and describe an improved feature selection method based on Segment-Boost. Segment-Boost learns the weak classifiers for the feature sets of various sizes through given training examples. Hence, our proposed feature selection method can consider discriminations and complementarities of features for selecting features. In addition, the proposed feature selection method requires lower computational cost than other boosting-based feature selection methods. For experiments, we extracted Gabor feature vectors from 400 face images of ORL facial database through Gabor filters. And randomly generated intra/extra-personal feature vectors from the Gabor feature vectors were used for training and testing examples. From the experimental results, we verified that Segment-Boost can select more discriminative features at lower computational cost than other boosting-based feature selection methods.