Efficient Wavelet Entropy-Based Face Recognition System

In the presented paper, the face images modeling in different circumstances such as occlusion, poses, and illumination was investigated. A novel research based on a new combination of the discrete wavelet transform (DWT) and wavelet packet (WP) methods in conjunction with log energy and sure entropies is proposed. Probabilistic neural network (PNN) was used for classification. The performance of the proposed method was investigated using a collection of experiments with different challenges ORL face database and extended Yale face database was conducted. For testing the proposed face recognition system on the ORL database, many classification scores were calculated for different training/testing sets. The comparative results obtained on the ORL database for different published methods based on the efficiency were performed and showed that our method was superior with rate reached 98.5% and relatively small, elapsed time about 4.4 sec was found. For the extended Yale face database B, the recognition rate was averaged at each time for 42 iterations. Furthermore, the recognition rate was calculated for different training/testing sets and excellent results were achieved. The contribution of the study was to use wavelet entropy for the face recognition task by finding a best combination of the method that leads to achieving a competent form guarantying better results. The major contribution of this study is that it is possible to use just a few real face images to model the illumination conditions in the extended Yale face database B.