Neural network-based fuzzy observer with application to facial analysis

Human facial wrinkles can be effectively utilized for facial analysis. It i s not an easy task, however, to extract features of wrinkledness directly f rom a camera image of a human face. For one thing, it is difficult to const ruct appropriate mathematical models for wrinkle features. In this work, a fuzzy observer is proposed as a means of providing linguistic descriptions about the image of a human face with wrinkles. In the proposed observer, so me well-defined classical image features and numerical information are tran sformed into fuzzy numbers. A feedforward multilayered artificial neural ne twork (ANN) is employed for parameter adjustment of the fuzzy observer base d on the available crisp-input fuzzy-output sample sets. An experiment is p erformed to demonstrate that the facial wrinkles can be indirectly estimate d by the proposed fuzzy observer. (C) 2000 Elsevier Science B.V. All rights reserved.