An HMM-Based Face Recognition Model under Variable Pose in Videos

We propose a model for extracting facial features robustly for face recognition under large pose variations in videos. The facial features are retrieved via Gabor Wavelet Transform with an embedded Hidden Markov Model (HMM), which decodes each observed face image into a state sequence. While an HMM can segment images into features at a fixed pose, multiple HMMs are trained for each individual to extract features robustly under large pose variation. The effectiveness of the proposed approach is validated through using the Sheffield Face Database. Our experiment shows better result than several other methods such as DCT+HMM,DWT+HMM, etc.