Personal Recognition Using ICA

Independent Component Analysis (ICA) has been recently used to find representation of images with neurophysiological plausibility. Here we extended it to the problem of extract features suitable for personal identification from both face images and speech signal. A two-channel biometric system is presented in this paper. Both the face recognition module and voice recognition subsystem of it are built on the features extracted by ICA. Those two channels are integrated using a weighted geometric average assuming that face features and voice features are independent. Preliminary experimental results demonstrate a success of ICA in the application of biometric feature extraction. The integrated system overcomes the limitations of an identification system solely on faces or speeches and also gets improvement in performance.