Spatial Merging for Face Detection

Face detection is a useful technique in computer vision. Many face detectors have been developed in the literature. These detectors to evaluate a face likelihood of a given sub-window. The sub-window must be scanned through an input image to detect faces. Since the detectors evaluate the scanned sub-windows independently, non-faces with high face likelihood are often misdetected. In this paper, we propose a novel face detection algorithm which explicitly uses difference of face likelihood distribution between faces and non-faces. The proposed algorithm can correctly classify the non-faces misdetected by the existing algorithm. The face likelihood distribution is generated and integrated to emphasize the difference between faces and non-faces. Experiments with pre-scanned data set and real-world images show that the proposed algorithm improves the detection rate approximately 20% and 10%, respectively