Face recognition based on modified discriminant independent component analysis

Independent component analysis (ICA) is a multivariable statistical analysis method which can be applied for face recognition problem. The aim of recognition is to approximately estimate the components from the raw image. Components plays an important role in face recognition systems. Consequently, these components are used for extraction of face image features. However, these features may not be appropriate for classification, since the ICA method does not consider the class information. For the purpose of optimizing the performance of ICA, the discriminant ICA (dICA) method, which is a combination of ICA and LDA methods, is utilized for face recognition in this study. We have also proposed particle swarm optimization method to improve the dICA performance, in which PSO is used instead of the gradient approach for learning dICA. The results of PSO-dICA method confirm our idea in classification experiments compared to other methods. Using proposed method on Yale B dataset, gives an average classification accuracy of 92.169% compared with an accuracy of 91.322% using when dICA and accuracy of 89.77% compared with ICA and accuracy of 86.18% using PCA and also accuracy of 84.76% using LDA.