Dear Jun,
You have misread that legend. It's not the case that the window of interest was determined based on an uncorrected statistical test. Panel (b) shows a TF image thresholded at p<0.01 uncorrected for illustrative purposes and panel (c) shows the results of corrected statistical test restricted to a window determined a-priori i.e. not based on the uncorrected test results.
Restricting the window of interest based on the same data is a common mistake and leads to false positives. You might be interested in looking at the new guidelines paper in Neuroimage
http://www.ncbi.nlm.nih.gov/pubmed/23046981 where these issues are discussed in the 'Statistics' chapter.
You can restrict the window of interest using a mask or 'small volume' button. When doing a cluster level correction you can note the size of the smallest significant cluster and the largest non-significant cluster and then display the results again using some number in between as the extent threshold. Then only significant clusters will be displayed.
Best,
Vladimir