As already mentioned, initial detection is greatly improved thanks to the mask obtained from the eigenbackground model. Since the mask corresponds to the are as which changed position, only these are checked for eye candidates. An example of this process is presented in Figure 10.