ISFER home page
     

    Mouth Segmentation Using Active Contours

    AUTHORS:
    P. Juran, P. Krsek, L.J.M. Rothkrantz, V. Hlavac

    KEYWORDS:

    ABSTRACT:
        This report describes our approach to localisation of the exact position of lips and teeth on a picture of mouth. Acquired information about mouth position will be used as input data for emotion recognition system. Basic demand is a good functionality on pictures of face without special make-up. Our algorithm consists of four steps:

    1. Preprocessing - image filtering
    2. Finding mouth-through line - valley detection and area growing
    3. Locating lips contour - active contours (snakes)
    4. Searching for teeth - active contours (snakes)
        In the first part of our report we are interested in image preprocessing. Colour information is used to obtain a picture with better contrast and the image is further processed by filtering. Then we locate the mouth-through line using two stage algorithm of valley detection and area growing. The line is used for definition of the starting conditions of the next step, the snakes. In the next part, we pay attention especially to definition of the energy of the active contour, terminating condition and the requirements of correct development. We experimented with several image definitions and their availability will be also discussed. In the final part we will describe results and conditions for realised experiments.
    -*-

    This page was created on January 19, 1998 by Anna Wojdel