Learnt combination in object detector ensembles

This paper considers the application of meta-classification techniques to the task of object detection. The technique of learnt combination is presented and evaluated in the domain of face detection: in the case of an ensemble of greyscale detectors it is shown to be as effective as pre-existing ensemble detection techniques. It is shown that considering detectors operating on different colour channels can improve performance; to exploit this, a method for selecting appropriate colour channels is also presented. Also presented is a new data set for training and evaluating colour-based face detectors.