Gender Classification by LUT Based Boosting of Overlapping Block Patterns

The paper addresses the problem of gender classification from face images. For feature extraction, we propose discrete Overlapping Block Patterns (OBP), which capture the characteristic structure from the image at various scales. Using integral images, these features can be computed in constant time. The feature extraction at multiple scales results in a high dimensionality and feature redundancy. Therefore, we apply a boosting algorithm for feature selection and classification. LookUp Tables (LUT) are utilized as weak classifiers, which are appropriate to the discrete nature of the OBP features. The experiments are performed on two publicly available data sets, Labeled Faces in the Wild (LFW) and MOBIO. The results demonstrate that Local Binary Pattern (LBP) features with LUT boosting outperform the commonly used block-histogram-based LBP approaches and that OBP features gain over Multi-Block LBP (MB-LBP) features.