Skip to Main Content
In human face detection applications, faces most frequently occupy a minor part of images. Therefore, preliminary segmentation of images into regions that contain "nonface" objects and regions that may contain "face" objects can greatly accelerate the process of human face detection thanks to a substantial narrowing of the area of search for the face detection algorithm. In the paper, an efficient and fast algorithm of such segmentation is suggested. The algorithm is tested on CMU B-set face database, which includes 153 faces in 23 grayscale images. The quality of many of the images is very low, the background part ("nonface" regions) is complex and the scenes are of different illumination conditions. The testing results show that the algorithm reduces the area of search for the final face detection to less than 1% of the image area while missing only 2 faces from 153.