A new model for pattern recognition

The hidden Markov model (HMM) has recently achieved impressive success in the field of pattern recognition, but some limitations and drawbacks restrict its performance. In this study, a new simple model is proposed to overcome the restrictions of HMM with a high reduction in the computational complexity. The training algorithm of the proposed model is built without iterations. It depends on the number of occurrences of each symbol in the training array, in case the discrete data form is used. However, for the continuous form, it uses the mean and covariance of the training data. On the other hand, the log-likelihood and the Mahalanobis distance are employed in the testing algorithm for indicating the highest matching between the testing data and the training parameters. This new model has been tested for face recognition; the experiments exhibit its advantage in terms of memory usage and processing time requirements, as well as recognition rate.