Implementation and enhancement of GMM face recognition systems using flatness measure

We propose a method of performance enhancement using the flatness measure (FM) for the Gaussian mixture model (GMM) face recognition systems. Also, we propose an implementation method of the face recognition systems to improve robustness by training the GMM, which is made by modeling the local features of the face images statistically, using an expectation maximization (EM) algorithm. The calculation burden is decreased using the flatness measure, and this improvement leads to the possibility of the real time face recognition. Experimental results show that the best rate of the face recognition is 100% for the cases of higher mixture and higher feature vector dimensions, though the rates depend much on the degree of mixture. We prove that the flatness measure is a useful pre-processing in the GMM-based face recognition systems by showing the improvement in performance by about 9% when the flatness measure is used even for the lower feature vector dimensions. 2-D DCT coefficients are used for the face feature vectors, and the experiments are carried out on the Olivetti Research Laboratory (ORL) face database.