Detecting gravitational wave background generated during inflation
through B-mode polarization of the cosmic microwave background (CMB) is
one of the main scientific goals of future CMB experiments. However, it
has already become clear that the synchrotron and thermal dust
emissions from our galaxy hinder the cosmological B-mode signal across
the sky. Here, assuming only Gaussianity and statistical isotropy and
homogeneity of the CMB map, we develop a new template cleaning method
taking into account spatial variations of spectral indices of the
foreground emissions, and apply it to various foreground models using
Monte-Carlo simulation technique. We find that the method successfully
remove the unwanted bias in determining the tensor-to-scalar ratio $r$
that was found in the simple template cleaning method presented in
Katayama and Komatsu 2011. In this poster, we present how small $r$ we
will be able to reach according to the frequency band and noise
specification proposed for the LiteBird experiment.