Skip to Main Content
Support vector machines are applied to to the problem of face detection using a feature-based approach. The specific feature focused on here is the cross-section of a nose. This focus is motivated by the unique "signature" of a nose, found consistently in a variety of images containing faces. The support vector classifier developed here makes use of a database of actual consumer images, provided by the Eastman Kodak Company. Use of this database ensures that the classifier will generalize to realistic images. An overall method incorporating a pre-processor, a support vector machine, and a post-processor is described. The method is demonstrated on a variety of consumer images, and statistical measures of performance are provided. A discussion is given on incorporating the proposed method into an overall face detection scheme.