Human-face recognition using neural network with mosaic pattern

A flexible robust human-face recognition method using neural networks with mosaic pattern is presented. In human-face recognition differences in facial expressions makes it difficult. However an appropriate mosaic face keeps the fundamental feature of the face. In the paper typical faces with different expressions are used as teacher patterns. As a result of employing such teacher patterns the property of the robustness of the neural network human-face recognition becomes very large and a neural network which is not so sensitive to human expression can be constructed. First it is shown that the presented neural network works well for differences of facial expressions such as smiling, crying, and irritated faces. Furthermore facing, right rotated, and left rotated faces are examined. A high recognition rate is attained for forty persons. The utilized network is a backpropagation neural one with three layers.