At risk of being inconsistent wrt older results (which were comparable), I have made alternations to the algorithm, which, based on my observations, have the potential of boosting recognition accuracy. Preliminary results might be ready soon.