Automatic face recognition based on skin masking and improved HMM

A new hidden Markov module (HMM) based face recognition system is presented in this paper. Face images were extracted automatically from live videos captured by a Creative WebCam, and faces were recognized in real time using an improved HMM face recognition algorithm. A fast face detection algorithm using boosted Haar-like features was applied initially to detect faces regions from the video stream. The detected face region was further refined by a skin color masking module to achieve more accurate face position. To improve the accuracy of HMM based face recognition algorithm, discrete wavelet transform, instead of discrete cosine transform, was used to extract observation sequences for HMM. Experiments were conducted using two face databases: The ORL database and the Nottingham color face image database. The results on both databases show that the proposed method can improve the accuracy by more than six percents. Further improvement has been observed when a skin masking module is used to refine the detected face region.