On Improving Face Detection Performance by Modelling Contextual Information

In this paper we present a new method to enhance object detection by removing false alarms and merging multiple detections in a principled way with few parameters. The method models the output of an object classifier which we consider as the context. A hierarchical model is built using the detection distribution around a target sub-window to discriminate between false alarms and true detections. Next the context is used to iteratively refine the detections. Finally the detections are clustered using the Adaptive Mean Shift algorithm. The specific case of face detection is chosen for this work as it is a mature field of research. We report better results than several baseline methods on the MIT+CMU, WEB and CINEMA face databases. We significantly reduce the number of false alarms while keeping the detection rate at approximately the same level and in certain conditions we recover misaligned detections.