Locating the head boundary with 2D continuous wavelet transform

The paper presents an approach to detect a human head boundary in a plain background image using a 2D Morlet continuous wavelet transform. A wavelet transform can be used as a powerful tool to extract singularities of images that, however, has not been well emphasized in the area of human face recognition. The 2D continuous wavelet transform based on the Euclidean group has an additional capability that can provide directional features of images. We utilize the rich edge information provided by a continuous wavelet in various directions to detect the head boundary, and then fuse that information to get the final result. Effectiveness of the proposed method is shown by tests with the Yale Face Database