Infrared Face Recognition System Using Cross Entropy Error Function Based Ensemble Backpropagation Neural Networks

Abstract

Abstract-This paper presents the development of cross-entropy error function based ensemble back-propagation neural network and its application as the classifier for frontal face recognition system. As the usually used quadratic error function has a drawback on its local minima problem, the used of a linear cross-entropy error function might directly found the global minima and achieve higher recognition capability. More over, we derive the cross-entropy error function for the negative corelation ensemble backpropagation and compare its characteristics with that of usually used quadratic based ensemble backpropagation neural networks. Experiments on face recognition system are conducted using infrared images, in order to obtain a robust face recognition for security puposes. In these experiments, face images with various expressions were utilized. Results show that the cross-entropy error function based ensemble back-propagation learning system performed better than that of quadratic error function based ensemble back-propagation learning system. It is clearly seen that the cross-entropy error function based ensemble backpropagation system could achieved higher recognition rates and more stable along various data sets, compare with that of quadratic error function based ensemble backpropagation neural network.Index Terms-Face recognition system, infrared images, ensemble backpropagation neural networks, cross-entropy error function.

Motion Symmetry Evaluation Using Accelerometers and Energy Distribution

“…The coefficients of the mathematical model were then evaluated so as to minimize the cross-entropy errors [29,[53][54][55], which heavily penalized extremely inaccurate outputs during the learning process. The efficiency of the classification was then evaluated by its accuracy and the cross-validation errors by the leave-one-out method.…”