Precise glasses detection algorithm for face with in-plane rotation

This paper proposes an accurate glasses detection algorithm for in-plane rotated faces. Firstly, to normalize the face with in-plane rotation to be an upright one, this paper proposes a new set of +/- 26.565A degrees haar-like features which can be calculated quickly to represent the features of rotated faces. With these new features, we address to build different rotated detectors by rotating upright face detector directly so as to achieve in-plane rotated face detection. Secondly, according to the rule that the objects with fixed and complex structure in a simple background are much easily to be detected, as the eyes have much complex structure in the face, we propose to detect the eyes for minimizing the glasses searching region. Because the nosepiece region of the glasses has much fixed structure and simple background, we use this part to train a new detector with the haar-like features and AdaBoost algorithm for precise glasses detection. This approach accomplishes glasses detection for in-plane rotated face with high speed and precision. The experimental results on the CMU-MIT, BioID, and CAS-PEAL-R1 face databases demonstrate that the algorithms for face, eye and glasses detection yield accurate results and maintain the speed advantage.