Improve Pattern Recognition Performance Based on Fractal Geometry Selection


In n-tuple and Hidden Markov Model(HMM) the recognition has been based on the feature selection. The feature selection in n-tuple depends on the number of tuples and its location. While, in HMM the feature has been related to the states. Where, the suitable features selection lead to optimal recognition. In this paper, a novel approach has presented for n-tuple and Hidden Markov model feature selection by using the Sierpiński fractal technique. The memory size and the recalling time taken to get individual classifier response has been reduced by 29.35% while the recognition is advancing the conventional n-tuple by 12.5% and 11.6% with and without frequency of occurrence respectively. In addition, the improvements noted in the HMMF proposed algorithm is 2.19% in recognition side, while it is 60% in complexity reduction. This approach is found to be robust in the presence of noise, where, the n-tupleF has advanced in recognition by 38.27% the conventional n-tuple algorithms, while HMMF has overperformed the n-tupleF by 14.44%. Simulation results show the maximum recognition is 92.3% for n-tupleF for character recognition, and HMMF is 99.98% for face recognition.