Effective face pose classification method

In this paper we present a novel appearance based approach to the problem of face pose classification. This method suggests the subject-independent pose classification of face images using bilateral filtering and wavelet transform as preprocessing and isometric projection based subspace learning for the extracting of discriminant feature vectors. Our proposed method is evaluated on a large image set of about 5000 face poses, drawn from the standard FERET database with inherent in-plane rotations and some manually added translations. The training set consists of 500 images equally distributed among frontal, approximately 22.5" rotated left and right quarter profile, and approximately 67.5" rotated left and right half profile and similarly the test set consists of 4500 images equally distributed among the five different types of poses. Experimental results show the effectiveness of our proposed method document.