Face detection and segmentation on a hierarchical image representation

This paper presents a face segmentation technique that works on a hierarchical region-based representation of the image. The algorithm bases its analysis strategy on a reduced set of regions that represent the image content at different scales of resolution. For each region, a set of simple one-class classifiers that rely on different shape, color and texture attributes is evaluated. The outputs of the classifiers are combined into a final face likelihood. The proposed system has been tested on a large set of images, providing very good results both in detection rate and accuracy of the segmented faces.