International Journal of Computer Science Issues

More than a traditional journal...

Face and Lip recognition has been benchmark problems in the field of biometrics and image processing. Various Artificial neural networks have been used for recognition purpose. This paper attempts at improving the recognition result for individuals by fusion of facial parts and lip using modular artificial neural network which employs parallel local experts with combinatory recognition techniques. Principal Component Analysis (PCA) and Regularized-Linear Discriminant Analysis (R-LDA) algorithm are used to extract low dimensional feature vector of images to drive neural networks effectively. Backpropagation Neural Network (BPNN) and Radial Basis Function Neural Network (RBFNN) are used as training algorithm for the database. Grimace database is used in this paper for carrying out the proposed methodology. Each facial image is divided into three sub image and a lip image. The three facial parts and the lip part are trained and tested individually. The fusion technique is applied using modular neural network by grouping sub images and the lip image in different network modules. Separate results obtained from each module are integrated to get the final result from the methodology used. This result set is compared with the result set obtained by training the sub images and the lip image individually. From the empirical and results finding it can be seen that the proposed methodology performs out better result.