Seven State HMM-Based Face Recognition System along with SVD Coefficients

Abstract

Face Recognition is a significant research area since it has plenty of application domains in pattern recognition, image processing, biometrics etc. Researchers contributed lot of algorithms and techniques to uncover the mask of face recognition arena. In our research, a Hidden Markov Model (HMM) based face recognition system along with Singular Value Decomposition (SVD) coefficients is discussed. Human face is divided into seven facial regions and a few quantized SVD Coefficients were trained to choose the facial features. Median Filtering is used as a preprocessing operation for efficient computation. Observation vectors are generated by dividing each face image into overlapping blocks and SVD coefficients acts as a base for constructing the observation sequence. Due to the discrete nature of HMM, quantization process is introduced to model the continuous observation vectors. The system is trained and tested on ORL face database consist of 400 images of 40 persons in Portable Gray Map (.pgm) format. Five face images of a person are considered for training the system and tested against 200 unseen faces. Our system achieves a recognition rate of 99.5% with a computational speed of 0.22 seconds per image. Choosing SVD coefficients as features increases the efficiency, in turn reducing the complexity. Experimental results revealed that our proposed system outperforms many of the traditional HMM based face recognition methods.