Improved KICA alogrithm and its application on face recognition

In order to reduce the original algorithm´s dependence on the primary matrix and raise the classification accuracy, we combine the five order Newton method and the steepest descent method. We also introduce punishing factors into algorithm, and apply them to the core of the iterative process of the Fast ICA algorithm, an improved kernel independent component analysis algorithm (KICA) is proposed in this paper. The improved algorithm is applied to face feature extraction. The traditional KICA algorithm and the improved algorithm are respectively used to extract features. Through the experiment of the ORL face database. We prove that the improved KICA algorithm is good for the separation of the independent component, and its classification accuracy is improved.