Ridge and valley based face detection

Abstract

Face detection is a challenging problem in face processing and recognition systems due to uncontrolled image acquisition conditions e.g., illumination, pose, etc. Many approaches for this problem have been proposed among which the most efficient could be Paul Viola's face detector using cascade of weak classifiers in Ada-boost learning with Harr features. Other face detectors use mainly Neural Network, Support Vector Machine, Hidden Markov Model, etc. Most of above approaches are robust and yielded relative high accuracy. However, these detectors depend upon sole pixel's intensities therefore it is not straightforward to reach high accuracy in face recognition, the process that requires conceptual description of individual face structures. In this paper, we introduce a new approach for face detection based upon Ridges and Valleys. These features can be extracted from face images and can be used to represent face structures at conceptual level. Initial results showed promising path to high accuracy in face recognition. Furthermore, this approach can be extended to the problem for flexible object detection and recognition.